# US 10,503,981 B2 · Method and apparatus for determining similarity of objects in images > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,503,981 B2 - Document ID: US10503981B2 - Status: Granted - Kind: B2 - Application number: 15/634,721 - Filed: 2017-06-27 - Published: 2018-12-27 - Granted: 2019-12-10 - Earliest claimed: 2017-06-27 - Assignees: Canon Kabushiki Kaisha - Applicants: Canon Kabushiki Kaisha - Inventors: Getian Ye; Ka Ming Leung - Web page: https://patentgrep.com/patent/US10503981B2 ## CPC - G06V 10/764: using classification, e.g. of video objects - G06F 18/2411: based on the proximity to a decision surface, e.g. support vector machines - G06V 20/52: Surveillance or monitoring of activities, e.g. for recognising suspicious objects (recognising microscopic objects G06V20/69) - G06V 40/103: {Static body considered as a whole, e.g. static pedestrian or occupant recognition} - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract A method of determining similarity of objects in images. Feature vectors are determined for objects in images captured by cameras operating in a training domain. Feature vectors are determined for the objects in images captured by cameras operating in a target domain, the cameras of the target domain operating with different environmental factors to the cameras of the training domain. A mapping is determined for a difference in the feature vectors of the training domain and the target domain. The difference in the feature vectors of the training domain and the target domain is converted to a matching space by applying the determined mapping to the feature vectors of the training domain and the target domain. A classifier is determined using data associated with the feature vectors of the training domain in the matching space. A similarity of the objects in the images captured in the target domain is determined using a difference of feature vectors for the objects by applying the classifier to the feature vectors of the objects in the matching space. ## Claims ### Claim 1 A method to determine similarity of objects in images, the method comprising: determining feature vectors for objects in images captured by cameras operating in a training domain; determining feature vectors for objects in images captured by cameras operating in a target domain, wherein the cameras of the target domain operate with different environmental factors to the cameras of the training domain; determining a mapping for a difference in the feature vectors of the training domain and the target domain; applying the determined mapping to the feature vectors of the training domain and the target domain to convert the difference in the feature vectors of the training domain and the target domain to a matching space; determining a classifier using data associated with the feature vectors of the training domain in the matching space; and determining a similarity of the objects in the images captured in the target domain using a difference of feature vectors for the objects by applying the classifier to the feature vectors of the objects in the matching space. ### Claim 2 (depends on 1) The method according to claim 1, further comprising determining a function of the difference between the feature vectors of the training domain and the target domain. ### Claim 3 (depends on 1) The method according to claim 1, further comprising determining an absolute difference between the feature vectors of the training domain and the target domain. ### Claim 4 (depends on 1) The method according to claim 1, wherein the difference in the feature vectors of the training domain and the target domain is a Euclidean distance. ### Claim 5 (depends on 1) The method according to claim 1, wherein the difference in the feature vectors of the training domain and the target domain is a cosine distance. ### Claim 6 (depends on 1) The method according to claim 1, further comprising determining a label for each of the determined differences. ### Claim 7 (depends on 1) The method according to claim 1, wherein determining the mapping includes using a composite kernel matrix to determine a projection. ### Claim 8 (depends on 1) The method according to claim 1, further comprising minimising difference in the feature vectors of the training domain and the target domain. ### Claim 9 (depends on 1) The method according to claim 1, further comprising representing a distribution of the difference in the feature vectors of the training domain and the target domain by a mean value of difference. ### Claim 10 (depends on 1) The method according to claim 1, wherein determining the mapping includes using an optimisation algorithm. ### Claim 11 (depends on 1) The method according to claim 1, wherein determining the classifier includes using support vector machines. ### Claim 12 (depends on 1) The method according to claim 1, wherein determining the classifier is based on the difference in the feature vectors of the training domain. ### Claim 13 (depends on 1) The method according to claim 1, further comprising generating, based on a result of determining the similarity of the objects in the images captured in the target domain, a response to either notify a user or to tag a gallery object, wherein the different environmental factors include images captured by different cameras, captured at different times, and images that represent different scenes. ### Claim 14 An apparatus to determine similarity of objects in images, the apparatus comprising: a determining unit configured to determine feature vectors for objects in images captured by cameras operating in a training domain; a determining unit configured to determine feature vectors for objects in images captured by cameras operating in a target domain, wherein the cameras of the target domain operate with different environmental factors to the cameras of the training domain; a determining unit configured to determine a mapping for a difference in the feature vectors of the training domain and the target domain; an applying unit configured to apply the determined mapping to the feature vectors of the training domain and the target domain to convert the difference in the feature vectors of the training domain and the target domain to a matching space; a determining unit configured to determine a classifier using data associated with the feature vectors of the training domain in the matching space; and a determining unit configured to determine a similarity of the objects in the images captured in the target domain using a difference of feature vectors for the objects by applying the classifier to the feature vectors of the objects in the matching space. ### Claim 15 A non-transitory computer-readable medium storing a program to cause a computer to execute a method to determine similarity of objects in images, the method comprising: determining feature vectors for objects in images captured by cameras operating in a training domain; determining feature vectors for objects in images captured by cameras operating in a target domain, wherein the cameras of the target domain operate with different environmental factors to the cameras of the training domain; determining a mapping for a difference in the feature vectors of the training domain and the target domain; applying the determined mapping to the feature vectors of the training domain and the target domain to convert the difference in the feature vectors of the training domain and the target domain to a matching space; determining a classifier using data associated with the feature vectors of the training domain in the matching space; and determining a similarity of the objects in the images captured in the target domain using a difference of feature vectors for the objects by applying the classifier to the feature vectors of the objects in the matching space. ### Claim 16 A system to determine similarity of objects in images, the system comprising: a memory for storing data and a computer program; a processor coupled to the memory for executing the computer program, the program having instructions for: determining feature vectors for objects in images captured by cameras operating in a training domain, determining feature vectors for objects in images captured by cameras operating in a target domain, wherein the cameras of the target domain operate with different environmental factors to the cameras of the training domain, determining a mapping for a difference in the feature vectors of the training domain and the target domain, applying the determined mapping to the feature vectors of the training domain and the target domain to convert the difference in the feature vectors of the training domain and the target domain to a matching space, determining a classifier using data associated with the feature vectors of the training domain in the matching space, and determining a similarity of the objects in the images captured in the target domain using a difference of feature vectors for the objects by applying the classifier to the feature vectors of the objects in the matching space.