Examples of the present disclosure relate to a method, apparatus, computer program and system for image analysis. According to certain examples, there is provided method comprising causing, at least in part, actions that result in: receiving orientation information of an image capturing device; receiving one or more features detected from an image captured by the image capturing device; and selecting a clustering model for clustering the features, wherein the clustering model is selected, at least in part, in dependence upon the orientation information.

One example provides a computing system comprising a storage machine storing instructions executable by a logic machine to extract features from a source and target images to form source and target feature maps, form a correlation map comprising a plurality of similarity scores, form an initial correspondence map comprising initial mappings between pixels of the source feature map and corresponding pixels of the target feature map, refine the initial correspondence map by, for each of one or more pixels of the source feature map, for each of a plurality of candidate correspondences, inputting a four-dimensional patch into a trained scoring function, the trained scoring function being configured to output a correctness score, and selecting a refined correspondence based at least upon the correctness scores, and output a refined correspondence map comprising a refined correspondence for each of the one or more pixels of the source feature map.

An edge computing system configured to dynamically offload tasks from a user device to an edge device. The edge computing system may receive a processed sensory feed from the user device, analyze the received processed sensory feed, and generate mapper output results. The edge computing system may compare the generated mapper output results to information received from a datastore, identify a correlation between a feature included in the received processed sensory feed and a feature included in the received information, and determine whether a confidence value associated with the identified correlation exceeds a threshold value. The edge computing system may further process the received processed sensory feed locally in the edge computing system or send the received processed sensory feed to a cloud component for further processing based on whether the confidence value exceeds the threshold value.

A method for detecting one or more topic transitions in a multimedia content includes identifying, one or more frames from a plurality of frames of the multimedia content based on a comparison between one or more content items in a first frame of the plurality of frames, and the one or more content items in a first set of frames of the plurality of frames. The method further includes determining at least a first score, and a second score for each of the one or more frames. Additionally, the method includes determining a likelihood for each of the one or more frames based at least on the first score, and the second score, wherein the likelihood is indicative of a topic transition among the one or more frames.

A system for place recognition is described herein. The system for place recognition comprises a plurality of sensors, a memory, and a processor. The memory is to store instructions and is communicatively coupled to the plurality of sensors. The processor is communicatively coupled to the plurality of sensors and the memory. When the processor is to execute the instructions, the processor is to detect features in a current frame and extract descriptors of the features of the current frame. The processor is also to generate a vocabulary tree using the descriptors and determine candidate key frames based on the vocabulary tree and detected features. The processor also is to perform place recognition via a first stage matching and a second stage matching.

Embodiments of the invention provide a method of visual saliency estimation comprising receiving an input sequence of image frames. Each image frame has one or more channels, and each channel has one or more pixels. The method further comprises, for each channel of each image frame, generating corresponding neural spiking data based on a pixel intensity of each pixel of the channel, generating a corresponding multi-scale data structure based on the corresponding neural spiking data, and extracting a corresponding map of features from the corresponding multi-scale data structure. The multi-scale data structure comprises one or more data layers, wherein each data layer represents a spike representation of pixel intensities of a channel at a corresponding scale. The method further comprises encoding each map of features extracted as neural spikes.

Object recognition by point-of-sale camera systems is aided by first removing perspective distortion. Yet pose of the objectrelative to the systemdepends on actions of the operator, and is usually unknown. Multiple trial counter-distortions to remove perspective distortion can be attempted, but the number of such trials is limited by the frame rate of the camera systemwhich limits the available processing interval. One embodiment of the present technology examines historical image data to determine counter-distortions that statistically yield best object recognition results. Similarly, the system can analyze historical data to learn what sub-parts of captured imagery most likely enable object recognition. A set-cover strategy is desirably used. In some arrangements, the system identifies different counter-distortions, and image sub-parts, that work best with different clerk- and customer-operators of the system, and processes captured imagery accordingly. A great variety of other features and arrangements are also detailed.

Various embodiments enable an image recognition system reduce the number image match candidates before running a full-fledged pair-wise match on all image match candidates. In order to accomplish this, each inventory image can be assigned to a group. For example, a title for a book sold by an electronic marketplace could be available in multiple languages, in multiple bindings, and the book could be available in print, audio book, or electronic book. Each one of these variations could be associated with its own similarly looking inventory image, each of which could be returned as a valid match to a query image for the book. Accordingly, the inventory images for these variations could be assigned to a group for the book and, instead of geometrically processing an image for each variation, the image match system can process a single image representing all of the variations.

The present application discloses a vehicle searching method and device, which can perform the steps of: calculating an appearance similarity distance between a first image of a target vehicle and several second images containing the searched vehicle; selecting several images from the several second images as several third images; obtaining corresponding license plate features of license plate areas in the first image and each of the third images with a preset Siamese neural network model; calculating a license plate feature similarity distance between the first image and each of the third images according to license plate feature; calculating a visual similarity distance between the first image and each of the third images according to the appearance similarity distance and the license plate feature similarity distance; obtaining a the first search result of the target vehicle by arranging the several third images in an ascending order of the visual similarity distances. The solution provided by the present application is not limited by application scenes, and it also improves vehicle searching speed and accuracy while reducing requirements of hardware such as cameras that collect images of a vehicle and auxiliary devices.

The accuracy of an image matching and/or object identification process can be improved by utilizing a BCM network-based process that maintains higher order relationships between features in an image. A dataset of images can be converted to floating point vectors and then processed using a BCM-based approach. The resulting vectors can be stored as an image library for purposes of matching subsequently received images. When a match is located for a query image, for example, information associated with the matching image can be provided in order to help identify one or more objects in the received query image.