Abstract: Techniques are provided that include obtaining a vocabulary including a set of content indices that reference corresponding cells in a descriptor space based on an input set of descriptors. A plurality of local features of an image are identified based on the vocabulary, the local features being represented by a plurality of local descriptors. An associated visual word in the vocabulary is determined for each of the plurality of local descriptors. A plurality of global signatures for the image are generated based on the associated visual words, wherein some of the plurality of global signatures are generated using local descriptors corresponding to different cropped versions of the image, two or more of the different cropped versions of the image being centered at a same pixel location of the image, and an image recognition search is facilitated using the plurality of global signatures to search a document image dataset.

Abstract: Described is a novel method for feature extraction for automatic gait recognition. This method uses Multi-kernel Fuzzy-based Local Gabor Binary Pattern. From a captured gait video sequence, the gait period is determined then a gait energy image is constructed to represent the spatial-temporal variations during one motion cycle of the gait sequence. Then, each gait sequence is represented with a feature vector. The computation of this vector is conducted by first applying the 2D Gabor filter bank then encoding the variations in the Gabor magnitude using a multi-kernel fuzzy local binary pattern operator. Finally, gait classification is performed using a support vector machine.

Abstract: Surface image data of a non-defective sample substrate is acquired, and surface image data of a substrate to be inspected is acquired. Differences between gradation values are calculated for pixels of the surface image data of the substrate to be inspected and corresponding pixels of the surface image data of the sample substrate. A constant value is added to the difference between gradation values of each pixel. In the case where the value acquired by addition is in a predetermined allowable range, it is determined that there is no defect for the substrate to be inspected. In the case where the value acquired by addition is outside of the allowable range, it is determined that the substrate to be inspected is defective. A defect in appearance on the substrate to be inspected is detected based on a pixel of which the value is outside of the allowable range.

Abstract: Described is a novel method for feature extraction for automatic gait recognition. This method uses Multi-kernel Fuzzy-based Local Gabor Binary Pattern. From a captured gait video sequence, the gait period is determined then a gait energy image is constructed to represent the spatial-temporal variations during one motion cycle of the gait sequence. Then, each gait sequence is represented with a feature vector. The computation of this vector is conducted by first applying the 2D Gabor filter bank then encoding the variations in the Gabor magnitude using a multi-kernel fuzzy local binary pattern operator. Finally, gait classification is performed using a support vector machine.

Abstract: A method for estimating a pose of a first object in relation to a second object, the second object comprising a visual marker comprising a plurality of ellipses comprises capturing a video image of the visual marker with an image capture device on the first object, then pre-processing frames of the video image on a graphics processing unit. The method comprises detecting the visual marker by finding contours in the frames to identify the plurality of ellipses and determining that a pattern of the plurality of ellipses match a known pattern of the visual marker. Then the method comprises obtaining coordinates of two or more of the plurality of ellipses of the visual marker, estimating the pose of the first object in relation to the second object by inputting the coordinates of the plurality of ellipses into a pose estimation algorithm, and filtering results of the pose estimation algorithm.

Abstract: Embodiments of the present invention relate to systems and methods for improving the training of machine learning systems to recognize certain objects within a given image by supplementing an existing sparse set of real-world training images with a comparatively dense set of realistic training images. Embodiments may create such a dense set of realistic training images by training a machine learning translator with a convolutional autoencoder to translate a dense set of synthetic images of an object into more realistic training images. Embodiments may also create a dense set of realistic training images by training a generative adversarial network (“GAN”) to create realistic training images from a combination of the existing sparse set of real-world training images and either Gaussian noise, translated images, or synthetic images.

Abstract: A method and a signal processor for receiving a data stream comprising at least two distinct sets of encoded data, at least one set of which is relative to transient/stochastic components of a signal. Based at least in part on the distinct sets of encoded data, the signal processor decodes and reconstructs a corresponding rendition of signal for each set of the encoded data. The distinct sets of renditions of signal are then combined into a single rendition of reconstructed signal.

Abstract: Image-based features may be significantly correlated with click-through rates of images that depict a product, which may provide a more formal basis for the informal notion that good quality images will result in better click-through rates, as compared to poor quality images. Accordingly, an image assessment machine is configured to analyze image-based features to improve click-through rates for shopping search applications (e.g., a product search engine). Moreover, the image assessment machine may rank search results based on image quality factors and may notify sellers about low quality images. This may have the effect of improving the brand value for an online shopping website and accordingly have a positive long-term impact on the online shopping website.

Abstract: An image processing apparatus includes: an extractor that divides image data in block units and extracts a maximum value and a minimum value of each block from the compressed image data; an edge detector that detects an edge at a front end or a rear end of an image; a correction value calculator that corrects the maximum value, or the maximum value and the minimum value of each block positioned in a certain range from the front end or the rear end and calculates a maximum value and a minimum value after the correction; and a converter that determines a conversion formula of a pixel value by using the maximum value and the minimum value before the correction and the maximum value and the minimum value after the correction and converts each pixel value of the image data obtained by decompressing the compressed data so as to apply sweeping correction.

Abstract: A data capture component of a mobile device receives information for an identification of a data field in a physical document. The data capture component receives a video stream comprising a plurality of frames, wherein each frame comprises a portion of the physical document. A frame is selected from the plurality of frames in the video stream. One or more text regions in the frame are identified. Each of the identified text region(s) in the frame is processed to identify data of each of the identified text region(s) and to select data of one of the identified text region(s) that corresponds to a set of attributes associated with the data field. The selected data is then compared with data of text regions of a subsequent frame. If the data of the text regions of the subsequent frame is a closer match to the set of attributes, the selected data is updated. A display field is then provided with the selected data for presentation with the frame in a user interface.

Abstract: To facilitate setting of a parameter at the time of generating an inspection image from an image acquired by using a photometric stereo principle. A photometric processing part generates an inspection image based on a plurality of luminance images acquired by a camera. A display control part and a display part switch and display the luminance image and the inspection image, or simultaneously display these images. An inspection tool setting part adjusts a control parameter of the camera and a control parameter of an illumination apparatus. Further, when the control parameter is adjusted, the display control part updates the image being displayed on the display part to an image where the control parameter after the change has been reflected.

Abstract: A deep neural network and a method for recognizing and classifying a multimedia data as one of a plurality of pre-determined data classes with enhanced recognition and classification accuracy and efficiency are provided. The use of the side branch(es) (or sub-side branch(es), sub-sub-side branch(es), and so on) extending from the main branch (or side branch(es), sub-side branch(es), and so on), the sequential decision making mechanism, and the collaborating (fusing) decision making mechanism in a deep neural network would equip a deep neural network with the capability for fast forward inference so as to enhance recognition and classification accuracy and efficiency of the deep neural network.

Abstract: An image processing apparatus includes a decomposition unit configured to decompose a color image into at least two color component images, a first obtaining unit configured to obtain a skewness of a histogram corresponding to each of the color component images, and a determination unit configured to determine based on the obtained skewnesses associated with the color component images whether the color image is glossy.

Abstract: There is provided a method of computing a camera pose of a digital image, comprising: computing query-regions of a digital image, each query-region maps to training image region(s) of training image(s) by a 2D translation and/or a 2D scaling, each training image associated with a reference camera pose, each query-region associated with a center point and a computed weighted mask that weights the query-region pixels according to computed correlations with the corresponding training image region, mapping cloud points corresponding to pixels of matched training image region(s) to corresponding images pixels of the matched query-regions according to a statistically significant correlation requirement between the center point of the query-region and the matched training image region, and according to the computed weight mask, and computing the camera pose according to an aggregation of the camera poses, and the mapped cloud points and corresponding image pixels of the matched query-regions.

Abstract: The present disclosure relates to a vehicle recognition method and system. The vehicle recognition method comprises: receiving a first image; extracting a first vehicle brand recognition region and a first color recognition region of the first image; recognizing the extracted first vehicle brand recognition region by a trained vehicle brand deep learning model; recognizing the extracted first color recognition region; and recognizing the vehicle based on the recognition results of the first vehicle brand recognition region and the first color recognition region.

Abstract: Systems, apparatuses and methods may provide for technology that partitions a high dynamic range (HDR) image into a plurality of regions and determines, on a per region basis, a luminance level of the HDR image. Additionally, the technology may select, on the per image basis, a encoding amount for each region in the plurality of regions based on the luminance level.

Abstract: A method, a device, and a computer program product for identifying objects in image data. The device is enabled to, in response to determining that an object type of a first set of object types is represented in the image data, indicate that the object type is represented in captured image data. The device is further enabled to, in response to determining that no object type of the first set of object types is represented in the image data, determine whether the image data comprises an object characteristic of a specific one of a second set of object types. The device is further enabled to, in response to determining that the image data comprises an object characteristic of the specific one of the second set of object types, indicate that the image data comprises an object characteristic of the specific one of the second set of object types.

Abstract: An interactive image segmenting apparatus and method are provided. The image segmenting apparatus and corresponding method include a boundary detector, a condition generator, and a boundary modifier. The boundary detector is configured to detect a boundary from an image using an image segmentation process. The feedback receiver is configured to receive information about the detected boundary. The condition generator is configured to generate a constraint for the image segmentation process based on the information. The boundary modifier is configured to modify the detected boundary by applying the generated constraint to the image segmentation process.

Abstract: A method for real time video processing for retouching an object in a video is presented The method includes providing an object in the video steam, where the object is at least partially and at least occasionally presented in frames of the video. The method sets a degree of retouching and generates a list of at least one element of the object selected based on a request of retouching and the degree of retouching. The method detects the at least one element of the object in the video and parameters of the at least one element and calculates new parameters of the at least one element according to the degree of retouching. Characteristic points are detected for each of the at least one element of the object and a mesh is generated based on the characteristic points for each of the at least one element of the object. The at least one element of the object in the video is tracked by aligning the mesh for each of the at least one element with a position of the corresponding each of the at least one element.

Abstract: Plural pieces of reference data are stored for plural pixel arrangement patterns. Each piece of reference data indicates which of first-type edge pixels needs to be subjected to determination to determine whether an outside neighboring first-type pixel adjacent to the first-type edge pixel exists. A piece of reference data corresponding to a target pixel block is specified as target reference data. A reference-data-based judgment is executed on the target pixel block to perform the determination onto those first-type edge pixels that are indicated by the target reference data as need to be subjected to the determination. A correction is executed on the target pixel block to correct the type of at least one first-type pixel contained in the target pixel block into the second type in a case where a result of the reference-data-based judgment indicates that no outside neighboring first-type pixel exists with respect to the target pixel block.