Abstract: Substrings within strings, such as words within words, are identified based at least in part on recorded behavior of users that have submitted the strings or substrings as search queries. The behavior may relate to actions taken by the users upon having submitting the search queries. The actions may be actions taken in connection with an electronic marketplace, such as actions related to the consumption of items offered in the electronic marketplace. The identified strings and corresponding substrings are used in connection with processing search queries. The strings and substrings may be used to update a search index and/or to modify received search queries for processing.

Abstract: Systems and methods are provided for creating contour images that represent the contour of objects reflected in images, calculating contour histogram descriptors of the contour images, and classifying images based in part on the histogram descriptors of the contour images. For example, a contour image of an image is created. A radial-polar grid having a plurality of radial-polar bins is then positioned on the contour image. A contour histogram descriptor is created to include a number of bins that correspond to the radial-polar bins of the radial-polar grid, where the contents of the bins of the contour histogram descriptor represent the number of pixels of the contour image that are located in the corresponding radial-polar bins of the radial-polar grid. Images are classified at least based in part on comparisons between contour histogram descriptors of the images and contour histogram descriptors of training images.

Abstract: A method, system and computer program product for encoding an image is provided. The image that needs to be represented is represented in the form of a Gaussian pyramid which is a scale-space representation of the image and includes several pyramid images. The feature points in the pyramid images are identified and a specified number of feature points are selected. The orientations of the selected feature points are obtained by using a set of orientation calculating algorithms. A patch is extracted around the feature point in the pyramid images based on the orientations of the feature point and the sampling factor of the pyramid image. The boundary patches in the pyramid images are extracted by padding the pyramid images with extra pixels. The feature vectors of the extracted patches are defined. These feature vectors are normalized so that the components in the feature vectors are less than a threshold.

Abstract: A method, system and computer program product for matching images is provided. The images to be matched are represented by feature points and feature vectors and orientations associated with the feature points. First, putative correspondences are determined by using feature vectors. A subset of putative correspondences is selected and the topological equivalence of the subset is determined. The topologically equivalent subset of putative correspondences is used to establish a motion estimation model. An orientation consistency test is performed on the putative correspondences and the corresponding motion estimation transformation that is determined, to avoid an infeasible transformation. A coverage test is performed on the matches that satisfy orientation consistency test. The candidate matches that do not cover a significant portion of one of the images are rejected. The final match images are provided in the order of decreasing matching, in case of multiple images satisfying all the test requirements.

Abstract: The present invention provides a software facility for identifying the items most relevant to a current query based on items selected in connection with similar queries. In preferred embodiments of the invention, the facility receives a query specifying one or more query terms. In response, the facility generates a query result identifying a plurality of items that satisfy the query. The facility then produces a ranking value for at least a portion of the items identified in the query result by combining the relative frequencies with which users selected that item from the query results generated from queries specifying each of the terms specified by the query. The facility identifies as most relevant those items having the highest ranking values.

Abstract: Techniques for injecting synthetic electronic advertisements into a production electronic advertisement system to test the effectiveness of new categories of electronic advertisement are provided. Synthetic advertisement content may be provided to real users of the production system in order to determine the effectiveness of new electronic advertisements and/or new categories of electronic advertisements. The synthetic advertisements content appears to users to be regular production advertisement that may include navigational element such as a hypertext link and associated text and/or graphics to be displayed in the electronic interface. Statistics regarding the effectiveness of the new electronic advertisements and/or new categories are gathered and may be used to determine which new electronic advertisements and/or new categories of electronic advertisements are effective enough to include in the production system.

Abstract: User behavior data can be used with language-specific rule sets to generate stemming databases useful for such tasks as indexing and search query processing. The terms contained in user queries, as well as user behavior with respect to those queries or results returned for those queries, can be analyzed to determine a relative measure (e.g., relative frequency) of various forms of those terms. When generating a stemming database, language-specific rule sets can be used to determine appropriate stemming rules, and where more than one potential rule is identified the user behavior data can be used to select what is likely the appropriate rule, at least for the respective environment. Whitelists or other such components can be used to handle specific or irregular forms that do not follow the general rules or otherwise are exceptions that might not otherwise be processed correctly.

Abstract: Disclosed are various embodiments of systems, methods and computer readable mediums for matrix viewing. In one embodiment, among others, a system comprises at least one computing device and an application executable in the at least one computing device. The application comprises logic that searches and gathers a plurality of item representations based at least in part on a search query retrieved from a client computing device. The application further comprises logic that encodes a network page for rendering by the client computing device. The network page provides a matrix view of at least a portion of the item representations. Also, the matrix view includes a plurality of streams. At least one of the streams comprises a moving progression of the item representations across at least a portion of the network page.

Abstract: Shape-based search of a collection of content associated with one or more images of inventory items (“inventory images”) is enabled at least in part by associating the collection of content and/or its associated inventory images with representative refinement shapes. Inventory items may be grouped into categories and at least one refinement shape may be created for each of the categories. A refinement-shape hierarchy may be created by arranging the refinement shapes into parent and child refinement shapes. Inventory images may be associated to at least one of the refinement shapes of the refinement-shape hierarchy based at least in part on similarities between the refinement shapes and shapes of the inventory items reflected in the inventory images.

Abstract: The present technology relates to methods, systems and computer program products for detecting objects in images captured by a camera of a mobile device. The detection of objects (such as barcodes, QR codes, and text) in images can be based at least in part on image gradients of the image. The image can be divided into a plurality of regions, each having a dominant gradient direction. Based at least in part on the dominant gradient directions of the regions satisfying orientation thresholds, the regions can be identified as candidate or non-candidate regions for corresponding to a predetermined object. The candidate regions can be merged if they satisfy a connecting criterion. The merged regions can then be analyzed to determine if the merged regions satisfy a geometric property of the predetermined object, such as having a rectangular shape substantially similar to that of a barcode.

Abstract: Multiple paths of an index tree may be traversed to discover a set of content descriptors that are match candidates for a set of query descriptors. A size of the set of candidate content descriptors may be optimized, for example, to reduce false positive matching errors, query latencies and/or index tree traversal times, at least in part by determining a number of child nodes to traverse based at least in part on current traverse level and/or traverse neighborhood thresholds. Index trees for large content descriptor sets may be built in resource constrained environments with approximation and/or refining build techniques.

Abstract: Systems and methods are described for identifying item attributes that are likely to be useful to users in refining searches for items. In one embodiment, when a user submits a search query to search for items, all or a portion of the search query is used to look-up a previously-generated list of item attributes (and optionally associated attribute values) that are associated with the search query and are useful for refining the search. These item attributes, and any associated attribute values, are then presented to the user in a selectable form such that the user can interactively narrow the search.

Abstract: A set of search requests may be analyzed to detect fixed phrases suitable for inclusion in a search index. Sets of candidate phrases may be identified among the search requests. Fixed phrases may be detected among the candidate phrases using statistical techniques, for example, by identifying phrases having a relatively high pointwise mutual information (PMI) with respect to component keywords. Fixed phrase detection may include keyword and/or phrase clustering. Clusters may correspond to topics defined using a latent Dirichlet allocation (LDA) procedure. Fixed phrase detection may include identifying phrases having relatively high PMI within particular clusters.

Abstract: A facility for navigating within a body of data using one of a number of distinct browse graphs is described. Initially, a navigation request is received. Based upon information contained in the received navigation request, the facility selects one of the plurality of browse graphs. In response to user input, the facility browses the body of data using the selected browse graph. The browse graphs may each correspond to a collection of the body of data, such as a website.

Abstract: The present invention provides a method and system for image matching. The method includes receiving a query image at a query-server. Further, the method includes, sending a request to one or more image-matching servers of a set of distributed image-matching servers to conduct an image-search corresponding to the query image. Furthermore, the method includes receiving at the query-server, a list of identified matches from the image-matching servers corresponding to the query image, wherein the list of matches is identified based on the image-search at the image-matching servers. Moreover, the method includes selecting one or more matches from the list of identified matches based on a score corresponding to the identified matches.

Abstract: Various algorithms are presented that enable an image of a barcode, captured by a camera of an electronic device, to be analyzed and decoded for use in obtaining information about an object or item associated with the barcode. The algorithms can account for variations in focus of the image, as may result in blur due to movement or a lack of auto-focus features. An offline training phase can be used to generate a set of possible symbol appearance templates, where at least some of the templates reflect simulations for varying levels and types of noise in an image. A decoding phase compares these templates to a captured image and performs a sequence-wide optimization to determine the overall most likely sequence of digits.

Abstract: Processes such as image matching can be improved by utilizing a Polar Accumulated Signed Gradient (PASG), a feature descriptor vector that utilizes polar coordinates. Such an approach is efficient because it avoids the costly image rotation and resampling of conventional processes. Further, the gradients can be sampled over different regions in order to achieve effective weighting without analyzing all gradients and applying a weighting factor. Such efficiencies enable image matching to be performed on devices with limited computing capacity, such as smart phones and tablet computers.

Abstract: Techniques for content segment optimization are disclosed. In one embodiment, the method for content segment optimization comprises receiving a user selection of a content item available on a network resource, providing a segment of the content item to a user for accessing, recording user activity on the network resource subsequent to the user accessing the content item segment, determining a type of user activity with respect to the content item based on the recorded user activity, and assigning a score to the segment provided to the user. The assigned score corresponds to the determined type of user activity with respect to the content item. Scores for each segment of the content item provided to users are combined to select a representative segment for the content item. The user activity includes purchasing the item, renting the item, adding the item to wishlist, or providing a review of the item.

Abstract: Techniques for facilitating a shape-based search of content items are disclosed. In some situations, the techniques include receiving a request for search of content that includes an image of an item and a glyph input by the user on the image that indicates a shape of the first item, extracting features associated with the item from an area of the image on which the glyph is input; identifying a shape category of the first item based at least in part on the indicated shape of the glyph; and conducting a search for items having features that match the extracted features of the first item, the search being conducted among a plurality of items associated with the identified shape category. The retrieved search results may be returned to a user.

Abstract: A method, system and computer program product for encoding an image is provided. The image that needs to be represented is represented in the form of a Gaussian pyramid which is a scale-space representation of the image and includes several pyramid images. The feature points in the pyramid images are identified and a specified number of feature points are selected. The orientations of the selected feature points are obtained by using a set of orientation calculating algorithms. A patch is extracted around the feature point in the pyramid images based on the orientations of the feature point and the sampling factor of the pyramid image. The boundary patches in the pyramid images are extracted by padding the pyramid images with extra pixels. The feature vectors of the extracted patches are defined. These feature vectors are normalized so that the components in the feature vectors are less than a threshold.