Abstract: An image processing device includes a target pixel detector configured to detect a plurality of target pixels in which noise is generated among a plurality of pixels included in an image sensor. The image processing device also includes a target pixel corrector configured to change target pixel values, which are pixel values of the plurality of target pixels, by using average pixel values of neighboring pixels included in a preset range based on a position of each of the plurality of target pixels. The image processing device further includes a target pixel compensator configured to compensate for the target pixel values by using an accumulation value obtained by accumulating values corresponding to a decimal fraction part of the average pixel values.

Abstract: A system comprises an encoder configured to compress attribute information for a point cloud and/or a decoder configured to decompress compressed attribute information for the point cloud. Attribute values for at least one starting point are included in a compressed attribute information file and attribute correction values used to correct predicted attribute values are included in the compressed attribute information file. Attribute values are predicted based, at least in part, on attribute values of neighboring points and distances between a particular point for whom an attribute value is being predicted and the neighboring points. The predicted attribute values are compared to attribute values of a point cloud prior to compression to determine attribute correction values. A decoder follows a similar prediction process as an encoder and corrects predicted values using attribute correction values included in a compressed attribute information file.

Abstract: Methods, systems, and non-transitory computer readable media are disclosed for intelligently generating partially textured accessible images. In one or more embodiments, the disclosed systems generate or access a color texture map specific to a given color vision deficiency. For example, the disclosed systems generate a color texture map that divides a color space into one or more textured segment of colors and one or more complementary untextured segments of colors. The disclosed systems can utilize the color texture map to intelligently apply different textures to subsets of pixels that contribute to color vision deficiency confusion. For example, the disclosed system maps pixels from a color image to the color texture map to identify a first subset of pixels corresponding to the textured segment of colors. The disclosed system generates a partially textured accessible image by applying a texture to the first subset of pixels.

Abstract: Systems and methods for performing data analytics on sensitive data, such as deduplication are disclosed herein. A first request to perform a first data analytics operation for a set of sensitive data instances may be received. A first set of image representations for the set of sensitive data instances may be retrieved. A first sensitive data instance and a second sensitive data instance may be clustered into a first cluster. The first data analytics operation may be performed on the first cluster.

Abstract: This disclosure provides systems, methods, and devices for image processing that supports high dynamic range (HDR) photography. In some aspects, a method of generating a full-resolution HDR photograph with in-sensor zoom includes receiving first and second image data corresponding to first and second exposure captures of a scene. First and second full-resolution image frames may be generated from the first and second image data, which are subsequently processed with HDR fusion to obtain an output image frame with higher dynamic range than either the first or second image data. The first full-resolution image frame may be determined from both the first and second image data by compensating the second image data for differences between the first and second exposures. Other aspects and features are also claimed and described.

Abstract: An image processing method includes: receiving a processing instruction for an image; screening out a plurality of target images from a plurality of acquired images; determining a first feature value and a second feature value of the target area of each of the target images; determining a similarity between the first feature values of the target areas in every two target images according to all the determined first feature values, and determining a similarity between the second feature values of the target areas in every two target images according to all the determined second feature values; and, if a first target similarity is within a first preset range, and a second target similarity is within a second preset range, determining that a first image and a second image belong to the same image set.

Abstract: Examples are described for applying different settings for image capture to different portions of image data. For example, an image sensor can capture image data of a scene and can send the image data to an image signal processor (ISP) and a classification engine for processing. The classification engine can determine that a first object image region depicts a first category of object, and a second object image region depicts a second category of object. Different confidence regions of the image data can identify different degrees of confidence in the classifications. The ISP can generate an image by applying a different settings to the different portions of the image data. The different portions of the image data can be identified based on the object image regions and confidence regions.

Abstract: A system and method for encoding a digital image may include obtaining image data of the digital image, wherein the image data comprises a plurality of image pixels, selecting a plurality of sample pixels from the plurality of image pixels, clustering the plurality of sample pixels based on an influence function to generate a plurality of hue clusters, classifying the image pixels to the plurality of the hue clusters, and assigning hue values of the hue clusters to the plurality of the image pixels.

Abstract: An image processing method and an image processing device are provided. The image processing method includes: acquiring an input image; extracting first and second pixel groups in the input image, wherein the first pixel group comprises first pixels with different positions, and the second pixel group comprises second pixels with different positions, the positions of the first pixels are different from the positions of the second pixels, the number of the first pixels are equal to the number of the second pixels, and the position of each first pixel in the first pixel group corresponds to the position of a respective second pixel in the second pixel group; when a preset similarity condition is satisfied between the first pixel and the respective second pixel, determining a first processing result of the first pixel as a second processing result of the respective second pixel.

Abstract: The present disclosure discloses a detection method and a detection system. The detection method comprises: creating a plurality of template images based on a reference object, wherein the reference object includes a plurality of units, and the plurality of template images are unit images with different average grayscale values; calculating a first average grayscale value of a unit image to be detected; selecting a first template image from the plurality of template images based on the first average grayscale value, wherein a difference between an average grayscale value of the first template image and the first average grayscale value is smallest; performing color difference detection on the unit image to be detected based on the first template image.

Abstract: Embodiments of this disclosure include a method and an apparatus for processing image. The method may include obtaining a to-be-processed image and performing image semantic segmentation on the to-be-processed image to obtain a semantically-segmented image. The semantically-segmented image may include a target region and a non-target region obtained through the semantic segmentation. The method may further include performing pose recognition on the to-be-processed image, to obtain a pose-recognized image recognizing skeletal region. The method may further include fusing the target region and the non-target region of the semantically-segmented image with the skeletal region of the pose-recognized image, to obtain a trimap comprising foreground region, background region, and recognition region. The method may further include generating, according to the to-be-processed image and the trimap, a transparency mask image for separating image from the to-be-processed image.

Abstract: A harvesting machine includes a header configured to gather harvested material into the harvesting machine during a harvesting operation, a conveyance subsystem configured to convey the harvested material from the harvesting machine to a receiving vehicle during the harvesting operation, an image capture system comprising at least one optical sensor, and a control system configured to determine a position of the receiving vehicle relative to the harvesting machine, determine an image magnification factor based on the determined position, and display, on a display device, an image of a portion of the receiving vehicle based on the image magnification factor.

Abstract: Introduced here is a technique to create small compressed image files while preserving data quality upon decompression. Upon receiving an uncompressed data, such as an image, a video, an audio, and/or a structured data, a machine learning model identifies an object in the uncompressed data such as a house, a dog, a text, a distinct audio signal, a unique data pattern, etc. The identified object is compressed using a compression treatment optimized for the identified object. The identified object, either before or after the compression, is removed from the uncompressed data. The uncompressed data with the identified object removed is compressed using a standard compression treatment.

Abstract: Methods, systems and apparatus, including computer programs encoded on computer storage media for determining asset efficiency. Unmanned Aerial Vehicles (UAVs) may be used to obtain aerial images of locations, property or structures. The aerial images may be geo-rectified, and a ortho-mosaic, digital surface model, or a point cloud may be created. In the context of an operation where mobile assets are used, such as construction or earth moving equipment, location-based event information may be obtained. The location-based event information may be used to determine road segment conditions or road topology where problematic road conditions likely exist.

Abstract: A target detection method is provided for a computing device. The method includes obtaining a current frame and a previous key frame corresponding to the current frame in a video frame sequence, determining a flow feature map and a flow field between the previous key frame and the current frame, obtaining, in response to determining the current frame is a non-key frame according to the flow feature map, a key frame feature corresponding to the previous key frame, and performing affine transformation on the key frame feature according to the flow field to obtain an image feature corresponding to the current frame, and performing target detection on the current frame according to the image feature to obtain a target detection result of the current frame.

Abstract: A search engine is configured to determine a set of main products that match search criteria of a search request. A search binary mask is generated based on the search criteria indicated in the search request which identify one or more ancillary products. If the search binary mask matches a reference binary mask representing the requested ancillary products, the main products are selected for the response to the search request. The search engine also verifies a current availability of the ancillary products associated with the selected main products and returns at least one of the main products for which the current availability has been determined.

Abstract: An information processing apparatus (2000) analyzes a captured image (12) generated by a camera (10) to determine a motion of a person. The camera (10) is a camera that images a display place where an item is displayed. The information processing apparatus (2000) detects a reference position (24) from the captured image (12). The reference position (24) indicates a position of a hand of the person. The information processing apparatus (2000) decides an analysis target region (30) to be analyzed in the captured image (12) using the reference position (24). The information processing apparatus (2000) analyzes the analysis target region (30) to determine the motion of the person.

Abstract: In implementations of correcting dust and scratch artifacts in digital images, an artifact correction system receives a digital image that depicts a scene and includes a dust or scratch artifact. The artifact correction system generates, with a generator of a generative adversarial neural network (GAN), a feature map from the digital image that represents features of the dust or scratch artifact and features of the scene. A training system can train the generator adversarially to reduce visibility of dust and scratch artifacts in digital images against a discriminator, and train the discriminator to distinguish between reconstructed digital images generated by the generator and real-world digital images. The artifact correction system generates, from the feature map and with the generator, a reconstructed digital image that depicts the scene of the digital image and reduces visibility of the dust or scratch artifact of the digital image.

Abstract: Systems and methods of improving alignment in dense prediction neural networks are disclosed. A method includes identifying, at a computing system, an input data set and a label data set with one or more first parts of the input data set corresponding to a label. The computing system processes the input data set using a neural network to generate a predicted label data set that identifies one or more second parts of the input data set predicted to correspond to the label. The computing system determines an alignment result using the predicted label data set and the label data set and a transformation of the one or more first parts, including a shift, rotation, scaling, and/or deformation, based on the alignment result. The computing system computes a loss score using the transformation, label data and the predicted label data set and updates the neural network based on the loss score.

Abstract: Disclosed herein are an apparatus and method for monitoring a user based on multi-view face images. The apparatus includes memory in which at least one program is recorded and a processor for executing the program. The program may include a face detection unit for extracting face area images from respective user images captured from two or more different viewpoints, a down-conversion unit for generating at least one attribute-specific 2D image by mapping information about at least one attribute in the 3D space of the face area images onto a 2D UV space, and an analysis unit for generating user monitoring information by analyzing the at least one attribute-specific 2D image.

Abstract: Apparatus and method for determining a quality score for virtual video cameras. For example, one embodiment comprises: a region of interest (ROI) detector to detect regions of interest within a first image generated from a first physical camera (PCAM) positioned at first coordinates; virtual camera circuitry and/or logic to generate a second image positioned at the first coordinates; image comparison circuitry and/or logic to establish pixel-to-pixel correspondence between the first image and the second image; an image quality evaluator to determine a quality value for the second image by evaluating the second image in view of the first image.

Abstract: A thermal image processing method includes determining a blending ratio of a first equalization algorithm to a second equalization algorithm that are different from each other, based on a result of analyzing an original histogram of an original thermal image; generating a first corrected thermal image obtained by applying the first equalization algorithm and the second equalization algorithm to the original thermal image according to the blending ratio; adjusting the blending ratio by comparing a corrected histogram of the first corrected thermal image with the original histogram; and generating a second corrected thermal image obtained by applying the first equalization algorithm and the second equalization algorithm to the original thermal image according to the adjusted blending ratio.

Abstract: A head-counter device (100) comprising a digital camera (1) adapted to provide a first digital image (IM1) representative of a counting zone (STR) of persons, the first image defining a first horizontal dimension (N) and a first vertical dimension (M), and a cropping module (7) configured for: analyzing the first image (IM1) and identifying a noise area (PCR) according to at least one of the following features: pixel light intensity, pixel color and/or presence of predefined patterns, cropping the noise area (PCR) from the first image (IM1) to obtain a second image (IM2) without the noise area, the noise area (PCR) being a peripheral portion of the first image having said first horizontal dimension and having a second vertical dimension (M?DSK+SM) shorter than the first vertical dimension.

Abstract: An object of the present invention is to extract an area of a foreground object with high accuracy. The present invention is an image processing apparatus including: a target image acquisition unit configured to acquire a target image that is a target of extraction of a foreground area; a reference image acquisition unit configured to acquire a plurality of reference images including an image whose viewpoint is different from that of the target image; a conversion unit configured to convert a plurality of reference images acquired by the reference image acquisition unit based on a viewpoint corresponding to the target image; and an extraction unit configured to extract a foreground area of the target image by using data relating to a degree of coincidence of a plurality of reference images converted by the conversion unit.

Abstract: A method of determining ripeness of a growing mushroom involves obtaining a sequence of images of a growing mushroom over a period of time and, from the sequence of images, measuring one or more features of the growing mushroom to obtain temporal measurements for the one or more features. The temporal measurements indicate rates at which the one or more features are changing. The temporal measurements are compared to pre-determined ripeness functions to determine the ripeness of the mushroom. The mushroom is picked if the comparison indicates that the mushroom is ready to be harvested.

Abstract: A system and a method of performing an image matting on an image are provided. The method includes detecting, by an image processing system, one or more objects in the image; determining, by the image processing system, a confidence map associated with the image using one or more image segmentation techniques for each of the one or more objects; and generating, by the image processing system, a tri-map for each of the one or more objects in the image from the confidence map based on at least one of a size of each of the one or more objects in the image and a distance between a first pixel in the image and a second pixel in at least one of the one or more objects in the image, wherein the tri-map is used to perform the image matting.

Abstract: Embodiments of the present disclosure discloses a method and apparatus for rendering an image, an electronic device and a computer readable storage medium. An example implementation of the method includes: acquiring a real color parameter and a face skin color parameter in a real user image including a face image area, determining lighting parameter information based on the real color parameter and the face skin color parameter, acquiring a to-be-rendered image obtained based on the real user image, and generating a texture parameter based on the lighting parameter information, and using the texture parameter to perform lighting rendering on the to-be-rendered image to obtain a rendered image.

Abstract: Disclosed is a method for simulating the rendering of a make-up product on the face of a subject, using a database of reference images including, for each reference individual, images of the face with and without the makeup product. The method includes: acquiring an image of the subject face without makeup; processing the image to extract, for each spatial area of each spatial frequency range of the image, first color feature values of the spatial area; determining, among the database of reference images, reference individuals having, when wearing no makeup, color feature values similar to the first color feature values of the subject; determining, from the first color feature values of the subject, and from color feature values of the reference individuals with and without the makeup product, second color feature values; and generating a modified image of the subject face based on the second color feature values.

Abstract: A processing apparatus combines a plurality of images based on a plurality of object images formed on an imaging plane of an image sensor by a plurality of lens units and to generate a combined image, and includes at least one processor or circuit that serves as an acquisition task configured to acquire information on a center position of each of the plurality of object images on the imaging plane, information on a correspondence relationship between the center position and positions of the plurality of images in the combined image, and conversion information for converting a first coordinate system in the imaging plane into a second coordinate system in the combined image, the conversion information being generated based on a correction function for correcting the plurality of object images, and a processing task configured to generate the combined image using the conversion information.

Abstract: An apparatus for image conversion includes a processor configured to classify a reference region representing a predetermined feature into a shadowed region and an unshadowed region, the reference region being in an aerial image represented in RGB color space; determine a tone correction factor so that a difference between an average luminance of the shadowed region and an average luminance of the unshadowed region in the aerial image represented in predetermined color space to which the color space of the aerial image is converted from RGB color space is less than a difference between an average luminance of the shadowed region and an average luminance of the unshadowed region represented in RGB color space; correct tones of the aerial image with the tone correction factor; and convert the color space of the aerial image from RGB color space to the predetermined color space to generate a color conversion image.

Abstract: A method of detecting unusual motion is provided, including: determining features occurring during a fixed time period; grouping the features into first and second subsets of the fixed time period; grouping the features in each of the first and second subsets into at least one pattern interval; and determining when an unusual event has occurred using at least one of the pattern intervals.

Abstract: A method of tracking an animal in a population of animals is disclosed that includes applying a visual identifier to the animal, associating reference identifier data of the visual identifier to animal data of the animal, capturing a digital image of the visual identifier applied to the animal, recognizing the visual identifier by digitally processing the captured digital image, and identifying the animal upon recognition of the visual identifier.

Abstract: Systems and methods for generating object segmentations across videos are provided. An example system can enable an annotator to identify objects within a first image frame of a video sequence by clicking anywhere within the object. The system processes the first image frame and a second, subsequent, image frame to assign each pixel of the second image frame to one of the objects identified in the first image frame or the background. The system refines the resulting object masks for the second image frame using a recurrent attention module based on contextual features extracted from the second image frame. The system receives additional user input for the second image frame and uses the input, in combination with the object masks for the second image frame, to determine object masks for a third, subsequent, image frame in the video sequence. The process is repeated for each image in the video sequence.

Abstract: A method for measuring an antenna downtilt angle based on a multi-scale deep semantic segmentation network is disclosed, including: collecting base station antenna data by using an unmanned aerial vehicle, and labeling an acquired antenna image with a labeling tool to make a data set; calling the data set for training and debugging a model; recognizing and detecting a target antenna, performing semantic segmentation on an output image, finally obtaining a target image finally segmented, and calculating a downtilt angle of the target image. The method is highly applicable, cost-effective, and safe.

Abstract: The computer stores the plurality of first positions in the image captured by the image pickup device and the second positions corresponding to the plurality of first positions in the graphic corresponding to the image as a relationship. The computer acquires a third position related to the object in the image, a positional relationship between a part of the plurality of first positions and the third position, and one of the relationships. A method specifies a position in a figure corresponding to a position in an image including generating a fourth position in the figure corresponding to the third position using parts or all. In a method, a part of the plurality of first positions is three first positions closest to the third position among the plurality of first positions. A method includes generating the fourth position using the sides of a triangle composed of the three first positions.

Abstract: Methods and systems disclosed herein relate generally to systems and methods for modifying pixel values of an image to improve the visibility of target pixel patterns. A pixel-simulation application accesses an initial image including an initial set of pixel values. The initial set of pixel values define, in an initial color space, a particular color of pixels that indicate a target pixel pattern. The pixel-simulation application generates, based on the initial set of pixel values, a simulated image including a modified set of pixel values that visually indicate another color of pixels in an intermediate color space. The pixel-simulation application generates a pixel map by identifying a difference between the initial set pixel values of the initial image and the modified set of pixel values of simulated image. The pixel-simulation application generates, for display, an output image based at least in part on the pixel map.

Abstract: One or more embodiments of the disclosure include systems and methods that use meta-learning to learn how to optimally find a new neural network architecture for a task using past architectures that were optimized for other tasks, including for example tasks associated with autonomous, semi-autonomous, assisted, or other driving applications. A computer implemented method of the disclosure includes configuring a search space lattice comprising nodes representing operator choices, edges, and a maximum depth. The method includes defining an objective function. The method further includes configuring a graph network over the search space lattice to predict edge weights over the search space lattice.

Abstract: Occlusion of facial features may be detected and assessed in an image captured by a camera on a device. Landmark heat maps may be used to estimate the location of landmarks such as the eyes, mouth, and nose of a user's face in the captured image. An occlusion heat map may also be generated for the captured image. The occlusion heat map may include values representing the amount of occlusion in regions of the face. The estimated locations of the eyes, mouth, and nose may be used in combination with the occlusion heat map to assess occlusion scores for the landmarks. The occlusion scores for the landmarks may be used control one or more operations of the device.

Abstract: Systems and methods for panoptic image segmentation are disclosed herein. One embodiment performs semantic segmentation and object detection on an input image, wherein the object detection generates a plurality of bounding boxes associated with an object in the input image; selects a query bounding box from among the plurality of bounding boxes; maps at least one of the bounding boxes in the plurality of bounding boxes other than the query bounding box to the query bounding box based on similarity between the at least one of the bounding boxes and the query bounding box to generate a mask assignment for the object, the mask assignment defining a contour of the object; compares the mask assignment with results of the semantic segmentation to produce a refined mask assignment for the object; and outputs a panoptic segmentation of the input image that includes the refined mask assignment for the object.

Abstract: A method for processing an image is presented. The method locates a subject and an object of a subject-object interaction in the image. The method determines relative weights of the subject, the object, and a context region for classification. The method further classifies the subject-object interaction based on a classification of a weighted representation of the subject, a weighted representation of the object, and a weighted representation of the context region.

Abstract: Banding effects in an image or video are assessed. The image or each frame of a video is decomposed into luminance or color channels. Signal activity is computed at each spatial location in each channel. A significance of the signal activity at each spatial location is determined by comparing each element of the signal activity with a significance threshold. Banding pixels are detected as those pixels of the image or frame having significant signal activity at their respective spatial locations and having non-significant signal activity of at least a minimum threshold percentage of neighboring pixels to the respective spatial locations.

Abstract: The present disclosure provides an interactive image matting method, a computer readable memory medium, and a computer device. The interactive image matting method includes steps: obtaining an original image; collecting foreground sample points on a hair edge foreground region of the original image and collecting background sample points on a hair edge background region of the original image by a human-computer interaction method to correspondingly obtain a foreground sample space and a background sample space; receiving a marking operation instruction input by a user, and smearing a hair region of the original image according to the marking operation instruction to mark unknown regions; traversing the unknown regions to obtain a pixel of each unknown region, traversing all the sample pairs to select a sample pair with a minimum overall cost function value for the pixel of each unknown region.

Abstract: A processing apparatus combines a plurality of images based on a plurality of object images formed on an imaging plane of an image sensor by a plurality of lens units and to generate a combined image, and includes at least one processor or circuit that serves as an acquisition task configured to acquire information on a center position of each of the plurality of object images on the imaging plane, information on a correspondence relationship between the center position and positions of the plurality of images in the combined image, and conversion information for converting a first coordinate system in the imaging plane into a second coordinate system in the combined image, the conversion information being generated based on a correction function for correcting the plurality of object images, and a processing task configured to generate the combined image using the conversion information.

Abstract: A method, system and computer program product for segmenting generic foreground objects in images and videos. For segmenting generic foreground objects in videos, an appearance stream of an image in a video frame is processed using a first deep neural network. Furthermore, a motion stream of an optical flow image in the video frame is processed using a second deep neural network. The appearance and motion streams are then joined to combine complementary appearance and motion information to perform segmentation of generic objects in the video frame. Generic foreground objects are segmented in images by training a convolutional deep neural network to estimate a likelihood that a pixel in an image belongs to a foreground object. After receiving the image, the likelihood that the pixel in the image is part of the foreground object as opposed to background is then determined using the trained convolutional deep neural network.

Abstract: A system and method for encoding a digital image may include obtaining image data of the digital image, wherein the image data comprises a plurality of image pixels, selecting a plurality of sample pixels from the plurality of image pixels, clustering the plurality of sample pixels based on an influence function to generate a plurality of hue clusters, classifying the image pixels to the plurality of the hue clusters, and assigning hue values of the hue clusters to the plurality of the image pixels.

Abstract: Embodiments are disclosed for identify a proportional dominant color palette of a vector graphics content and recolor the vector graphics content by editing the proportional dominant color palette. In particular, one method comprises obtaining a vector graphic, determining, by a palette extraction manager, a plurality of colors from the vector graphic, grouping, by the palette extraction manager, the plurality of colors into a plurality of groups based on color characteristics, determining, by the palette extraction manager, a plurality of dominant colors corresponding to the plurality of groups based on a subset of the plurality of colors associated with each group, determining, by the palette extraction manager, a plurality of weights corresponding to the plurality of colors, and generating, by the palette extraction manager, a proportional dominant color palette for the vector graphic based on the plurality of dominant colors and the corresponding plurality of weights.

Abstract: A method and a device for processing feature points of an image are provided. A specific embodiment of the method includes obtaining an image to be processed; determining weights of the feature points of the image to be processed to obtain a weight set; and according to the weights, selecting target numbered feature points as target feature points of the image to be processed. The weights include a texture weight; the texture weight and a color change scope of pixels in a target sized image region in which the feature points locate are directly proportional. The embodiment can reduce the number of feature points of the image, and further release the storage pressure of feature points regarding the image.

Abstract: Integrating an image capturing mechanism into an ecommerce application is described. Images of items offered for sale via ecommerce channels associated with an ecommerce service provider and performance metrics associated with the images and/or content items with which the images are presented via the ecommerce channels can be analyzed. Based partly on the analyzing, a characteristic associated with the images that are associated with a performance metric that satisfies a threshold or that are presented by the content items that are associated with the performance metric that satisfies the threshold can be determined. A new image of an item to be offered for sale by a merchant via an ecommerce channel can be received and, based at least partly on the characteristic, it can be determined whether to integrate the new image into a content item for the merchant, wherein the content item to be presented via the ecommerce channel.

Abstract: Signal detection and recognition employees coordinated illumination and capture of images under to facilitate extraction of a signal of interest. Pulsed illumination of different colors facilitates extraction of signals from color channels, as well as improved signal to noise ratio by combining signals of different color channels. The successive pulsing of different color illumination appears white to the user, yet facilitates signal detection, even for lower cost monochrome sensors, as in barcode scanning and other automatic identification equipment.

Abstract: An approach using 3D algorithms to solve 2D head localization problems is disclosed. A system can extrapolate aspects of one part of an object, e.g., extract characteristics of a person's head, using a 2D input image of another part of the object, e.g., a 2D image of the person's face. The system then selects an appropriate 3D model by the use of facial features detected in an image of a person's face. Using the selected 3D model and the 3D rotation angles provided by a face detector, the system rotates the model and then projects the model to a 2D shape. The system then scales and translates, e.g., transforms, the 2D shape to match the 2D face bounding box. Then, using the transformed 2D shape, the system extracts a bounding box for the extracted portion of an object, e.g., the head of the person depicted in the 2D input image.