Abstract: Methods, systems, and apparatus, including computer program products, for ranking search results for queries. The method includes calculating a visual similarity score for one or more pairs of images in a plurality of images based on visual features of images in each of the one or more pairs; building a graph of images by linking each of one or more images in the plurality of images to one or more nearest neighbor images based on the visual similarity scores; associating a respective score with each of one or more images in the graph based on data indicative of user behavior relative to the image as a search result for a query; and determining a new score for each of one or more images in the graph based on the respective score of the image, and the respective scores of one or more nearest neighbors to the image.

Abstract: A smart-focusing technique includes identifying an object of interest, such as a face, in a digital image. A focus-generic classifier chain is applied that is trained to match both focused and unfocused faces and/or data from a face tracking module is accepted. Multiple focus-specific classifier chains are applied, including a first chain trained to match substantially out of focus faces, and a second chain trained to match slightly out of focus faces. Focus position is rapidly adjusted using a MEMS component.

Abstract: Techniques for calibrating a classification system, wherein one or more objects in at least one video are classified, are provided. At least one view associated with the at least one video is obtained. The at least one view is partitioned into at least one region. A given object is classified in accordance with its location in reference to the at least one region. In an additional embodiment, one or more object models are obtained. At least one normalized size of the one or more objects is defined within at least one view associated with the at least one video in accordance with the one or more object models. The one or more objects are classified in accordance with the at least one defined normalized size.

Abstract: There is provided an image processing apparatus that analyzes a pathology image obtained by imaging an analysis object including a setting device for calculating an amount of a luminance difference characteristic as an amount of characteristic of each pixel of a pathology image, and for setting a cellular tissue region in the pathology image, based on the amount of the luminance difference characteristic, using a cellular tissue detector that has studied in advance. The cellular tissue detector may have studied in advance by statistical learning that uses an image corresponding to a cellular tissue.

Abstract: Methods for recognizing a category of an object are disclosed. In one embodiment, a method includes determining, by a processor, a preliminary category of a target object, the preliminary category having a confidence score associated therewith, and comparing the confidence score to a learning threshold. If the highest confidence score is less than the learning threshold, the method further includes estimating properties of the target object and generating a property score for one or more estimated properties, and searching a supplemental image collection for supplemental image data using the preliminary category and the one or more estimated properties. Robots programmed to recognize a category of an object by use of supplemental image data are also disclosed.

Abstract: An object detection apparatus that detects an object to be detected captured in a determination image according to a feature amount of the object to be detected preliminarily learned by the use of a learning image, the object detection apparatus including a detector 2 causing strong classifiers 21m to operate in order of lower classification accuracy, continuing processing when the strong classifier 21m has determined that the object to be detected is captured in the determination image, and determining that the object to be detected has not been detected without causing the strong classifier 21m having classification accuracy higher than the aforementioned strong classifier 21m to operate, when the strong classifier 21m has determined that the object to be detected is not captured in the determination image, wherein the strong classifier 21m inputs a classification result of the strong classifier 21m having classification accuracy lower than the aforementioned strong classifier 21m and determines whether the

Abstract: An image recognition information attaching apparatus includes a retrieving unit that retrieves image information on a per piece basis of identification information, from the image information having the identification information associated thereto in advance, a generator unit that generates feature information from the image information retrieved by the retrieving unit, and a learning unit that provides a learning result by learning a relation between the feature information generated by the generator unit and the identification information of the image information corresponding to the feature information, using a stochastic model including a mixture of a plurality of probability distributions.

Abstract: A method of computerized searching for visual objects on an electronic document is disclosed. After receiving a selection of a template visual object, a template raster image is generated therefrom. The template visual object is comprised of one or more template object components. A document raster image is generated from the document, which includes document object symbols comprised of one or more document object components. A broad match candidate subset of document object symbols is generated based upon comparisons to the template object symbol. Then, a narrowed match candidate set is generated from those document object symbols of the broad match candidate subset based upon a comparison of the template object component to the document object components.

Abstract: A training system, training method, and a system and method of use of a trained classification system are provided. A classifier may be trained with a first “cheap” view but not using a second “costly” view of each of the training samples, which is not available at test time. The two views of samples are each defined in a respective original feature space. An embedding function is learned for embedding at least the first view of the training samples into a common feature space in which the second view can also be embedded or is the same as the second view original feature space. Labeled training samples (first view only) for training the classifier are embedded into the common feature space using the learned embedding function. The trained classifier can be used to predict labels for test samples for which the first view has been embedded in the common feature space with the embedding function.

Abstract: A method of classifying pixels in an image is described that includes calculating for each target pixel in the image, a functional value based on a median value of a block of pixels including the target pixel and storing the functional value for each pixel. Pixels in the image are then analyzed to determine if they correspond to edges in the image and if so, are classified as edge pixels. Next the stored functional values are analyzed to define a flat area delimiting function for the image. The stored functional values that do not correspond to edge pixels are then analyzed to define an image detail delimiting function and the non-flat area pixels are classified as being either flat area pixels or detail pixels based on the flat area delimiting function and the detail delimiting function.

Abstract: In one embodiment, there is provided a method for an Optical Character Recognition (OCR) system. The method comprises: recognizing an input character based on a plurality of classifiers, wherein each classifier generates an output by comparing the input character with a plurality of trained patterns; grouping the plurality of classifiers based on a classifier grouping criterion; and combining the output of each of the plurality of classifiers based on the grouping.

Abstract: Techniques and structures are disclosed in which one or more distortion categories are identified for an image or video, and a quality of the image or video is determined based on the one or more distortion categories. The image or video may be of a natural scene, and may be of unknown provenance. Identifying a distortion category and/or determining a quality may be performed without any corresponding reference (e.g., undistorted) image or video. Identifying a distortion category may be performed using a distortion classifier. Quality may be determined with respect to a plurality of human opinion scores that correspond to a particular distortion category to which an image or video of unknown provenance is identified as belonging. Various statistical methods may be used in performing said identifying and said determining, including use of generalized Gaussian distribution density models and natural scene statistics.

Abstract: In one aspect, a method to reduce false alarms in identifying whether a candidate image is from an object class includes projecting the candidate image onto an object class subspace and projecting the candidate image onto a non-object class subspace. The method also includes determining whether the candidate image is from the object class using a Bayesian decision function based on the projections on the object class subspace and the non-object class subspace. In another aspect, a method to reduce false alarms in identifying whether a candidate mine image is from a mine class includes projecting the candidate mine image onto a mine subspace and projecting the candidate mine image onto a non-mine subspace. The method also includes determining whether the candidate mine image represents a mine using a Bayesian decision function based on the projections on the mine class subspace and the non-mine class subspace.

Abstract: A method of detecting a face in an image includes performing face detection within a first window of the image at a first location. A confidence level is obtained from the face detection indicating a probability of the image including a face at or in the vicinity of the first location. Face detection is then performed within a second window at a second location, wherein the second location is determined based on the confidence level.

Abstract: Bicycles present before a vehicle are identified. A picked-up image having intensity values that are in accordance with the temperatures of objects is acquired to extract image regions having intensity values representative of higher temperatures than a background. An object, which includes a pair of first object portions (legs) and a second object portion (a tire), would be identified as a bicycle if the pair of first object portions, which have greater dimensions in length in the vertical direction than in width in the horizontal direction and the distance between which is less than a predetermined value in the horizontal direction, are detected from the extracted image regions and further if the second object portion, which has a different intensity value from any one of the first object portions and also has a length greater than a predetermined value in the vertical direction, is existent between the pair of first object portions.

Abstract: A method for text recognition includes generating a number of text hypotheses for an image, for example, using an HMM based approach using fixed-width analysis features. For each text hypothesis, one or more segmentations are generated and scored at the segmental level, for example, according to character or character group segments of the text hypothesis. In some embodiments, multiple alternative segmentations are considered for each text hypothesis. In some examples, scores determined in generating the text hypothesis and the segmental score are combined to select an overall text recognition of the image.

Abstract: Embodiments include a scene classification system and method. In one embodiment, a method includes forming a first plurality of image features from an input image, processing the first plurality of image features in the first scene classifier.

Abstract: Method for recommending photographs chosen from a set of digital photographs, comprising: a tagging step for tagging the digital photographs according to determined objective parameters obtained from their own features and determined subjective parameters obtained from users' behaviour toward said photographs; a tuning step for obtaining a tuning score scoret for each photograph by weighing their objective and subjective parameters according to a relevant function; starting from the tuning score of each photograph, obtaining a total score scoreTOTAL for each photograph; provide the at least one user with the photographs recommendation according to the total score scoreTOTAL of each photograph.

Abstract: A bit code converter transforms a learning feature vector using a transformation matrix updated by a transformation matrix update unit, and converts the transformed learning feature vector into a bit code. When the transformation matrix update unit substitutes a substitution candidate for an element of the transformation matrix, a cost function calculator fixes the substitution candidate that minimizes a cost function as the element. The transformation matrix update unit selects the element while sequentially changing the elements, and the cost function calculator fixes the selected element every time the transformation matrix update unit selects the element, thereby finally fixing the optimum transformation matrix.

Abstract: A pattern recognition method, applicable to input information including a plurality of regions, includes obtaining a certainty at which each region of the input information includes a pattern, selecting one or more regions having a relatively high-level certainty among the plurality of regions, and performing pattern detection processing on the selected region or regions.

Abstract: A system for determining at least one characteristic of wood furnish from an upstream source. The system includes: an inclined panel comprising a transparent window having a top surface for the wood furnish to slide down; a lighting means adjacent the transparent window for lighting the wood furnish visible through a bottom surface of the window; an image capturing means adjacent the transparent window for capturing an image of the wood furnish visible through the bottom surface of the window; and a processing means in communication with the image capturing means for deriving from the captured images the at least one characteristic of the wood furnish.

Abstract: Statistical approaches to large-scale image annotation are described. Generally, the annotation technique includes compiling visual features and textual information from a number of images, hashing the images visual features, and clustering the images based on their hash values. An example system builds statistical language models from the clustered images and annotates the image by applying one of the statistical language models.

Abstract: A method of locating features of an object, of a class of objects, of a class of objects, within a target image. The method comprises initializing a set of feature points within the target image, each feature point corresponding to a predetermined feature for objects of the class of objects; deriving a set of template detectors, from the set of feature points, using a statistical model of the class of objects, each template detector comprising an area of image located about the location of a feature point for an object of the class of objects; comparing the set of template detectors with the target image; and updating the set of feature points within the target image in response to the result of the comparison.

Abstract: A correlation image detection method is provided that co-registers sonar images by finding peaks in correlation images. To obtain the peaks, the mean of the absolute values of the correlation coefficients in the correlation image is found and the Rayleigh parameter is determined from the mean. Based on the Rayleigh parameter, an appropriate threshold can be determined using a desired probability of false detection. The threshold can be chosen such that the probability of a single false detection over the expected life of the mission for which correlation detection is being performed is extremely low. The peak value in the image is determined and a correlation is considered detected when the peak value is greater than the product of the threshold and the Rayleigh parameter. If a detection occurs, the correlation image detector returns the transformation that co-registers the two images.

Abstract: A dual cross-media relevance model (DCMRM) is used for automatic image annotation. In contrast to the traditional relevance models which calculate the joint probability of words and images over a training image database, the DCMRM model estimates the joint probability by calculating the expectation over words in a predefined lexicon. The DCMRM model may be advantageous because a predefined lexicon potentially has better behavior than a training image database. The DCMRM model also takes advantage of content-based techniques and image search techniques to define the word-to-image and word-to-word relations involved in image annotation. Both relations can be estimated by using image search techniques on the web data as well as available training data.

Abstract: In image processing according to the prior art, the important part of photographic image data (referred to herein as the object) could not be determined and therefore required human participation. A computer 21 which is the core of image processing calculates an edginess which is an image variation from a differential value of data for adjacent picture elements in a step SA110, and determines object picture elements by selecting only images with a large variation in steps SA120, SA130. As optimum parameters for contrast correction and lightness compensation are calculated from image data for object picture elements in steps SA310-SA330, image processing indicators based on object picture elements are determined, and optimum image processing can be performed automatically.

Abstract: An apparatus capable of detecting location of object contained in image data and its detecting method are disclosed. The apparatus comprises an image capturing module, a weight assignment module, and a processing module. The image capturing module is for capturing an image. The weight assignment module performs the pixel weight/probability assignment according to the priori information and the image, and figures out the initial gravity center of the object according to the object location initialization. The processing module performs the statistical analysis according to the result of the pixel weight/probability assignment and the initial gravity center of the object so as to obtain the analysis result and update the object location. The processing module determines whether or not the analysis result meets the preset value, if it does, the processing module outputs an estimated result; if it doesn't, the processing module repeats the foregoing processes.

Abstract: An automated system and a method for extracting a region of interest from a digital image are disclosed. The method includes identifying a subset of training images from a larger set of training images, each training image in the set having a respective identified region of interest. The subset is identified based on a measure of similarity between the digital image and the images in the set of training images. At least one region of interest is extracted from the digital image based on an analysis of the identified regions of interest in the subset of training images.

Abstract: Two faces may be compared by calculating distances between different regions of the windows, and choosing one of the distances as the difference between the images. Two images are examined to detect the location of the face in the images. The faces may then be geometrically and photometrically rectified. A sliding window that is smaller than the whole face may be positioned at various locations over the images, and a descriptor is calculated for each window position. The descriptor for a window at one location in one image is compared with descriptors for windows in the neighborhood of that location in the other image. The lowest distance between window descriptors is chosen. The process is repeated for all window positions, resulting in a set of distances. The distances are sorted, and one of the distances is chosen to represent the difference between the two faces.

Abstract: An apparatus for providing pattern detection may include a processor. The processor may be configured to iteratively test different models and corresponding scales for each of the models. The models may be employed for modeling parameters corresponding to a visually detected data. The processor may be further configured to evaluate each of the models over a plurality of iterations based on a function evaluation of each of the models, select one of the models based on the function evaluation of the selected one of the models, and utilize the selected one of the models for fitting the data.

Abstract: A blur classification module may compute the probability that a given pixel in a digital image was blurred using a given two-dimensional blur kernel, and may store the computed probability in a blur classification probability matrix that stores probability values for all combinations of image pixels and the blur kernels in a set of likely blur kernels. Computing these probabilities may include computing a frequency power spectrum for windows into the digital image and/or for the likely blur kernels. The blur classification module may generate a coherent mapping between pixels of the digital image and respective blur states, or may perform a segmentation of the image into blurry and sharp regions, dependent on values stored in the matrix. Input image data may be pre-processed. Blur classification results may be employed in image editing operations to automatically target image subjects or background regions, or to estimate the depth of image elements.

Abstract: An image processing system provides faster than real-time skin detection and localization. The system uses the highly optimized architecture of a graphics processing unit to quickly and efficiently detect and locate skin in an image. By performing skin detection and localization on the graphics processing unit, the image processing system frees the main system processor to perform other important tasks, including running general purpose applications. The speed with which the image processing system detects and localizes skin also facilitates subsequent processing steps such as face detection and motion tracking.

Abstract: A method of making a selected-image collection including providing a set of relationships between an individual and a plurality of different persons, each relationship in the relationship set having a pre-determined relationship weight value; providing a collection of images, the image collection including images having at least two persons of the plurality of different persons present in the images; using a processor to select from the relationship set a relationship weight value corresponding to each person present in each image in the image collection; assigning an image weight value to each image in the image collection, the image weight value corresponding to a combination of the selected relationship weight value(s); and selecting images from the image collection based on the image weight values to make the selected-image collection.

Abstract: Techniques for calibrating a classification system, wherein one or more objects in at least one video are classified, are provided. At least one view associated with the at least one video is obtained. The at least one view is partitioned into at least one region. A given object is classified in accordance with its location in reference to the at least one region. In an additional embodiment, one or more object models are obtained. At least one normalized size of the one or more objects is defined within at least one view associated with the at least one video in accordance with the one or more object models. The one or more objects are classified in accordance with the at least one defined normalized size.

Abstract: A method, system and computer readable media for real-time chromakey matting using image statistics. To identify the chroma key spectrum, the system/method executes in three stages. In an off-line training stage, the system performs semi-automatic calibration of the chroma key parameterization. In the real-time classification stage, the system estimates the alpha matte on a GPU. Finally, an optional error minimization stage improves the estimated matte, accounting for misclassifications and signal noise. Given the resulting matte, standard alpha blending composites the virtual scene with the video feed to create the illusion that both worlds coexist.

Abstract: Technologies are generally described for methods and systems for generating a representation of a skin condition. The method may include receiving first image data relating to a first image of a skin condition and second image data relating to a second image of the skin condition. The method may further include performing a component analysis of the first and second image data to generate a first and second weighted component analysis results for the first and second image data. The weighted component analysis results may represent the respective image data as a weighted sum of components. The method may further include performing a sampling analysis to the weighted component analysis results to generate a weighted image representation for the skin condition. The weighted image representation may represent the skin condition as a weighted sum of the components.

Abstract: An object recognition system in which fall of the recognition rate is suppressed when an object is recognized based on an image even if there is a partial concealment and the object can be recognized even if the region of concealment is large with large calculation amount. With regard to each of a plurality of partial regions of an object image, partial recognition score of recognition object category is determined by judging whether it is a recognition object category or not. Under a condition that it is a recognition object category, total score is calculated using the total product of nonoccurrence probability of the partial recognition score, and a judgment is made that the object is not a recognition object category by that total score.

Abstract: A system and method for generating an image representation are provided. The image is modeled as a set of mixture weights, one for each of a set of reference image models, such as Gaussian mixture models (GMMs). The weights are derived by optimizing an objective function in which each reference image model is associated with its respective weight.

Abstract: A cortex-like learning machine, called a probabilistic associative memory (PAM), is disclosed for recognizing spatial and temporal patterns. A PAM is usually a multilayer or recurrent network of processing units (PUs). Each PU expands subvectors of a feature vector input to the PU into orthogonal vectors, and generates a probability distribution of the label of said feature vector, using expansion correlation matrices, which can be adjusted in supervised or unsupervised learning by a Hebbian-type rule. The PU also converts the probability distribution into a ternary vector to be included in feature subvectors that are input to PUs in the same or other layers. A masking matrix in each PU eliminates effect of corrupted components in query feature subvectors and enables maximal generalization by said PU and thereby that by the PAM. PAMs with proper learning can recognize rotated, translated and scaled patterns and are functional models of the cortex.

Abstract: A method is provided for automatically providing a digital image rating of photo retouching. The method includes the step of receiving at a computer, including a processor, a first set of pixel data of an original image and a second set of pixel data of a retouched image. The method also includes using the processor to determine a plurality of geometric statistics and a plurality of photometric statistics from the first and second sets of pixel data. The method further includes the step of using the processor to quantify a rating of the retouched image based upon the geometric statistics and photometric statistics to indicate deviation of the retouched image from the original image. A system is also provided to perform the steps.

Abstract: A blur classification module may compute the probability that a given pixel in a digital image was blurred using a given two-dimensional blur kernel, and may store the computed probability in a blur classification probability matrix that stores probability values for all combinations of image pixels and the blur kernels in a set of likely blur kernels. Computing these probabilities may include computing a frequency power spectrum for windows into the digital image and/or for the likely blur kernels. The blur classification module may generate a coherent mapping between pixels of the digital image and respective blur states, or may perform a segmentation of the image into blurry and sharp regions, dependent on values stored in the matrix. Input image data may be pre-processed. Blur classification results may be employed in image editing operations to automatically target image subjects or background regions, or to estimate the depth of image elements.

Abstract: A system and method for estimating a set of landmarks for a large image ensemble employs only a small number of manually labeled images from the ensemble and avoids labor-intensive and error-prone object detection, tracking and alignment learning task limitations associated with manual image labeling techniques. A semi-supervised least squares congealing approach is employed to minimize an objective function defined on both labeled and unlabeled images. A shape model is learned on-line to constrain the landmark configuration. A partitioning strategy allows coarse-to-fine landmark estimation.

Abstract: A set of ordered characters is received in association with information specifying the locations of the characters within the image of the document. Language-conditional character probabilities for each character are determined based on a set of language models and the ordering of the characters. Neighbor characters associated with a target character are identified based on the locations of the characters. Language-conditional character probabilities associated with the neighbor characters and language-conditional character probabilities associated with the target character are combined to generate a local language-conditional likelihood associated with the target character, the local language-conditional likelihood representing a concordance of the target character to a language model.

Abstract: Described is a system for rapid directed area search utilizing particle swarm optimization. The system first extracts salient regions from an input image. The system then detects regions of interest from the salient regions utilizing particle swarm optimization, wherein a swarm of software agents, or particles, cooperate to locate an objective function optima, or region of interest, in an image. A set of local feature descriptors are then extracted from the image, wherein a local feature descriptor corresponds to a neighborhood surrounding a point of interest in a region of interest in the image. Additionally, the set of local feature descriptors are clustered hierarchically into a database so that a closest match between a new input image and a stored image can be determined. Finally, the matching regions of the two images are registered to align matching regions to allow detection of changes between the images.

Abstract: An image processing apparatus is provided which includes a characteristic quantity calculation unit that calculates a characteristic quantity for each pixel of image data from the image data, a tightness determination unit that determines tightness indicating a degree of dispersion of pixels in image data for tightness determination by extracting the pixels having a same characteristic quantity by using the characteristic quantity and generating the image data for tightness determination composed of the pixels having the same characteristic quantity, and a pixel set calculation unit that calculates a set of the pixels from the pixels in the image data for tightness determination in accordance with a determination of the tightness.

Abstract: An information processing apparatus includes a characteristic amount calculating unit calculating a characteristic amount for each of a plurality of n different image patterns, a specifying unit specifying a best-matching image pattern among the plurality of n image patterns for each of frames forming a learning moving picture and having temporal continuity, a computing unit computing a collocation probability Pij indicating a probability that, for a frame located at a position where a temporal distance to a frame for which a first image pattern Xi is specified among the plurality of n image patterns is within a predetermined threshold ?, a second image pattern Xj is specified among the plurality of n image patterns, and a grouping unit grouping the plurality of n image patterns by using the computed collocation probability Pij.

Abstract: A system and method for determining statistical data within an image having pixels with relatively higher bit depth per band and a system and method of converting a first image within the image having pixels with the relatively higher bit depth per band into a second image having a relatively lower bit depth per band. The higher bit depth per band includes greater than 8 bits per band. The method of determining statistical data for pixels of an image includes dividing the image having pixels with the higher bit depth per band into a plurality of polygonal sections, each polygonal section including substantially homogenous features; computing statistical data for pixels within each polygonal section; and storing the statistical data with the image.

Abstract: An album creating apparatus for crating an album by automatically selecting an appropriate image to be laid out in an image layout frame arranged in a template and laying out the selected image in the image layout frame is provided.

Abstract: An apparatus for providing pattern detection may include a processor. The processor may be configured to iteratively test different models and corresponding scales for each of the models. The models may be employed for modeling parameters corresponding to a visually detected data. The processor may be further configured to evaluate each of the models over a plurality of iterations based on a function evaluation of each of the models, select one of the models based on the function evaluation of the selected one of the models, and utilize the selected one of the models for fitting the data.

Abstract: A method for providing face pose estimation for face detection may include utilizing a selected portion of classifiers in detectors to determine coarse pose information for a candidate face in an image, determining fine pose information for the candidate face based at least in part on the determined coarse pose information, and employing another portion of the classifiers in the detectors to perform face detection based at least in part on the fine pose information to determine whether the candidate face corresponds to a face. An apparatus and computer program product corresponding to the method are also provided.