摘要:
To compute a signature for an object comprising or represented by a set of vectors in a vector space of dimensionality D, statistics are computed that are indicative of distribution of the vectors of the set of vectors amongst a set of regions Ri, i=1, . . . , N of the vector space, at least some statistics associated with each region are binarized to generate sets of binary values ai, i=1, . . . , N indicative of statistics of the vectors of the set of vectors belonging to the respective regions Ri, i=1, . . . , N; and a vector set signature is defined for the set of vectors including the sets of binary values ai, i=1, . . . , N. The computing, binarizing, and defining operations may be repeated for two sets of vectors, and a quantitative comparison of the two sets of vectors determined based on the corresponding vector set signatures.
摘要:
To compute a signature for an object comprising or represented by a set of vectors in a vector space of dimensionality D, statistics are computed that are indicative of distribution of the vectors of the set of vectors amongst a set of regions Ri, i=1, . . . , N of the vector space, at least some statistics associated with each region are binarized to generate sets of binary values ai, i=1, . . . , N indicative of statistics of the vectors of the set of vectors belonging to the respective regions Ri, i=1, . . . , N; and a vector set signature is defined for the set of vectors including the sets of binary values ai, i=1, . . . , N. The computing, binarizing, and defining operations may be repeated for two sets of vectors, and a quantitative comparison of the two sets of vectors determined based on the corresponding vector set signatures.
摘要:
An instance-level retrieval method and system are provided. A representation of a query image is embedded in a multi-dimensional space using a learned projection. The projection is learned using category-labeled training data to optimize a classification rate on the training data. The joint learning of the projection and the classifiers improves the computation of similarity/distance between images by embedding them in a subspace where the similarity computation outputs more accurate results. An input query image can thus be used to retrieve similar instances in a database by computing the comparison measure in the embedding space.
摘要:
A method and system are disclosed for conducting text-based searches of images using a visual signature associated with each image. A measure of string similarity between a query and an annotation associated with each entry in a first database is computed, and based upon the computed string similarity measures, a set of entries from the first database is selected. Each entry of the first database also includes an associated visual signature. At least one entry is then retrieved from a second database based upon a measure of visual similarity between a visual signature of each of the entries in the second database and the visual signatures of the entries in the selected set. Information corresponding to the retrieved entries from the second database is then generated.
摘要:
A system and method for computing a place profile are disclosed. The method includes providing a geographical definition of a place, retrieving a set of images based on the geographical place definition. With a classifier, image-level statistics for the retrieved images are generated. The classifier has been trained to generate image-level statistics for a finite set of classes, such as different activities. The image-level statistics are aggregated to generate a place profile for the defined place which may be displayed to a user who has provided information for generating the geographical definition or used in an application such as a recommender system or to generate a personal profile for the user.
摘要:
As set forth herein, a computer-based method is employed to align a sequences of images. Metadata associated with images from two or more sources is received and a time stamp is extracted from the metadata. The images are sorted into sequences based at least in part upon the image source. The similarity of images from disparate sequences is measured and image pairs from disparate sequences with a similarity greater than a predetermined threshold are identified. A sequence of images is aligned by minimizing the misalignment of pairs.
摘要:
An automated image processing system and method are provided for class-based segmentation of a digital image. The method includes extracting a plurality of patches of an input image. For each patch, at least one feature is extracted. The feature may be a high level feature which is derived from the application of a generative model to a representation of low level feature(s) of the patch. For each patch, and for at least one object class from a set of object classes, a relevance score for the patch, based on the at least one feature, is computed. For at least some or all of the pixels of the image, a relevance score for the at least one object class based on the patch scores is computed. An object class is assigned to each of the pixels based on the computed relevance score for the at least one object class, allowing the image to be segmented and the segments labeled, based on object class.
摘要:
A method of maximizing a concave log-likelihood function comprises: selecting a pair of parameters from a plurality of adjustable parameters of a concave log-likelihood function; maximizing a value of the concave log-likelihood function respective to an adjustment value to generate an optimal adjustment value, wherein the value of one member of the selected pair of parameters is increased by the adjustment value and the value of the other member of the selected pair of parameters is decreased by the adjustment value; updating values of the plurality of adjustable parameters by increasing the value of the one member of the selected pair of parameters by the optimized adjustment value and decreasing the value of the other member of the selected pair of parameters by the optimized adjustment value; and repeating the selecting, maximizing, and updating for different pairs of parameters to identify optimized values of the plurality of adjustable parameters.
摘要:
Object clustering techniques are disclosed. A nonnegative sparse similarity matrix is constructed for a set of objects. Nonnegative factorization of the nonnegative sparse similarity matrix is performed. Objects of the set of objects are allocated to clusters based on factor matrices generated by the nonnegative factorization of the nonnegative sparse similarity matrix.
摘要:
An image adjustment includes adapting a universal palette to generate (i) an input image palette statistically representative of pixels of an input image and (ii) a reference image palette statistically representative of pixels of a reference image, and adjusting at least some pixels of the input image to generate adjusted pixels that are statistically represented by the reference image palette. In some embodiments, a user interface for controlling the image adjustment includes a display and at least one user input device, the user interface displaying a set of colors indicative of the regions of color space represented by a palette and receiving a selection of one or more regions of the color space, so that the image adjustment adjusts those pixels of the input image lying within the one or more selected regions of the color space.