SYSTEM FOR AUTOMATED SEGMENTATION OF IMAGES THROUGH LAYOUT CLASSIFICATION

• US 20150363660A1
• Filed: 06/11/2015
• Published: 12/17/2015
• Est. Priority Date: 06/12/2014
• Status: Abandoned Application

• Current Assignee: Asap54.com Ltd.
• Original Assignee: Asap54.com Ltd.
• Inventors: VIDAL, ANDRE, HEESCH, DANIEL
• Application Number:
  US14/737,467
• Publication Number:
  US 20150363660A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/532 Query formulation, e.g. graphical querying
• CPC Class Codes: G06F 16/5854 using shape and object relationship
• CPC Class Codes: G06F 18/23213 with fixed number of clusters, e.g. K-means clustering
• CPC Class Codes: G06F 18/24137 Distances to cluster centroïds
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/759 Region-based matching
• CPC Class Codes: G06V 10/763 Non-hierarchical techniques, e.g. based on statistics of modelling distributions
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects

Abstract

A system for extracting one or more regions of interest from a plurality of images to retrieve images based on visual similarity to a query image. Sequences of image processing associated with a segmentation strategy selected by user are performed on a set of training images to identify the region of interests. The segmentation strategy and the regions of interest are stored, as well as a visual signature of the image that captures the global layout of the image. New images are processed for which no segmentation strategy had previously been defined. A search is made through the layout signatures collected from the set of training images to identify images with similar layouts. Given a query and its visual characteristics, the system finds images stored in the database with visually similar regions of interest.

Claims

1. A processing system for manually selecting and combining image processing sequences to extract a region of interest from an image, comprising:

a server comprising a server processor and a server database;

a client device comprising a client processor, a client database and a display unit to display a user-interface;

the client processor is configured to load at least one image from the client database selected by an operator using the user-interface and to transmit said at least one image to the server processor for processing over a communications network;

the server processor applies a current sequence of image processing operations selected by the operator to said at least one image, stores a result of the current sequence of image processing operations applied on said at least one image in the server database, and transmits the result of the current sequence of image processing operations to the client device over the communications network;

the client processor, in response to the receipt of the result from the server, displays the result of the current sequence of the image processing operations applied on said at least one image on the display unit, after each display of the result, the client processor either (a) receives an acceptance of the result of the current sequence of the image processing operations from the operator via the user-interface and transmits the acceptance of the result of the current sequence of the image processing operations to the server over the communications network;

or (b) receives an adjustment to the current sequence of image processing operations from the operator via the user-interface, and transmits the adjustment to the current sequence of image processing operations to the server over the communications network for further processing by the server processor;

the server processor, in response to the receipt of the adjustment to the current sequence of image processing operations from the client device, stores the current sequence of image processing operations as a previous sequence of image processing operations in the server database, applies the adjustment to the current sequence of image operations to the image, stores a result of the adjustment to the current sequence of image operations applied to said at least one image in the server database, stores the adjustment to the current sequence of image processing operations as the current sequence of image processing operations, and transmits the result of the current sequence of image processing operations to the client device over the communications network; and

the server processor, in response to the receipt of the acceptance of the result of the current sequence of the image processing operations from the client device, associates and stores the current sequence of image processing operations as a segmentation strategy for said at least one image in the server database.

2. The processing system of claim 1, wherein the server processor automatically determines parameters of each image processing operation, receives an adjustment to one or more parameters of an image processing operation and applies the parameter adjustment to the image processing operation to the image.

3. The processing system of claim 1, wherein the server database comprises a plurality of images processed by the server processor and a segmentation strategy associated with each processed image;

wherein the server processor selects a set of reference images from the server database and transmits the set of reference images to the client device over the communications network; and

wherein the client processor receives an instruction to add a new image to or delete an image from the set of reference images from the user via the user interface, and transmits the instructions to the server over the communications network.

4. A retrieval system, comprising:

a communications network;

a server comprising a server processor, a classifier and a server database comprising a set of reference images processed by the processing system of claim 1 and a segmentation strategy associated with each reference image;

a plurality of user client devices, each comprising a client processor, a client database and a display unit to display a user-interface;

a user client device associated with a user transmits a set of search images to the server for processing over the communications network;

for each search image,

the server processor extracts a layout signature from each search image;

the classifier selects a candidate image from the set of reference images stored in the database with a layout signature similar to said each search image; and

the server processor applies a segmentation strategy associated with the candidate image to said each search image.

5. The retrieval system of claim 4, wherein the classifier clusters the reference images based on global image features;

and wherein the server processor stores each reference image processed in the server database by an associated segmentation strategy and a layout signature that captures a global layout of the image, the layout signature being a histogram of oriented gradients.

6. The retrieval system of claim 5, wherein the server processor stores each reference image processed in the server database by a category of a region of interest.

7. The retrieval system of claim 4, wherein the classifier determines the segmentation strategy for said each search image based on a similarity to the reference images in same category as said each search image.

8. The retrieval system of claim 4, wherein the classifier determines the candidate image based on k nearest neighbor search, k visually similar reference images are determined based on their layout signatures.

9. The retrieval system of claim 8, wherein the classifier groups images with same segmentation strategy into clusters, a centroid representing each group of images;

and wherein the classifier determines the k nearest neighbor by determining k closest centroids.

10. The retrieval system of claim 8, wherein the classifier determines the k nearest neighbors by employing at least one of the following:

locality sensitive hashing, vector approximation files, best-bin first, or balanced box-decomposition trees.

11. The retrieval system of claim 4, wherein the classifier identifies one or more regions of interest in said each search image based on the selected segmentation strategy;

and wherein the server processor determines visual descriptors for different perceptual dimensions of said one or more regions of interest in said each search image.

12. The retrieval system of claim 11, wherein the perceptual dimensions are color, shape and texture.

13. The retrieval system of claim 11, wherein the server processor employs a bag of words representation such that each visual descriptor is a histogram of visual words, each visual word corresponding to an aspect of the perceptual dimension.

14. The retrieval system of claim 11, wherein the server processor employs a cosine similarity measure to compute a similarity score based on two visual descriptors.

15. The retrieval system of claim 4, wherein the user client device associated with the user receives a query comprising a query image and optional search criteria from the user via the user-interface associated with the user client device;

wherein the client processor of the user client device transmits the query to the server over the communications network;

wherein the server processor extracts a layout signature from the query image;

wherein the classifier selects a candidate image from the set of reference images stored in the database with a layout signature similar to the query image; and

wherein the server processor applies a segmentation strategy associated with the candidate image to the query image.

16. The retrieval system of claim 15, wherein the classifier identifies one or more regions of interest in the query image based on the selected segmentation strategy;

and wherein the server processor determines visual descriptors for different perceptual dimensions of the regions of interest in the query image.

17. The retrieval system of claim 15, wherein the server processor computes visual descriptors on the regions of interest in the query image, determines one or more search images from the server database that are similar to the query image, and ranks the identified searched images based on relevance.

18. The retrieval system of claim 15, wherein the user client device associated with the user receives a uniform resource locator of the image selected by the user via the user-interface associated with the user client device;

and wherein the client processor of the user client device transmits the uniform resource locator the server for processing over the communications network.

19. The retrieval system of claim 15, wherein the user client device associated with the user receives a category selection as the optional search criteria from the user via the user-interface associated with the user client device;

wherein the client processor transmits the category selection to the server for processing over the communications network; and

wherein the classifier selects a segmentation strategy for the query image in accordance with the category selection.

20. The retrieval system of claim 4, wherein the user client device associated with the user receives a query comprising a query image and optional search criteria from the user via the user-interface associated with the user client device;

wherein the client processor of the user client device extracts a layout signature from the query image and transmits the query and the layout signature of the query image to the server over the communications network;

wherein the classifier selects a candidate image from the set of reference images stored in the database with a layout signature similar to the query image; and

wherein the server processor applies a segmentation strategy associated with the candidate image to the query image.