# US 8,542,905 B2 · Determining the uniqueness of a model for machine vision > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 8,542,905 B2 - Document ID: US8542905B2 - Status: Granted - Kind: B2 - Application number: 12/981,268 - Filed: 2010-12-29 - Published: 2012-07-05 - Granted: 2013-09-24 - Earliest claimed: 2010-12-29 - Assignees: Cognex Corporation - Applicants: Xiaoguang Wang; Lowell Jacobson - Inventors: Xiaoguang Wang; Lowell Jacobson - Web page: https://patentgrep.com/patent/US8542905B2 ## CPC - G06F 18/2193: {based on specific statistical tests} - G06F 18/28: Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Described are methods and apparatuses, including computer program products, for determining model uniqueness with a quality metric of a model of an object in a machine vision application. Determining uniqueness involves receiving a training image, generating a model of an object based on the training image, generating a modified training image based on the training image, determining a set of poses that represent possible instances of the model in the modified training image, and computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image. ## Claims ### Claim 1 A computerized method for determining a quality metric of a model of an object in a machine vision application, the method comprising: receiving a training image; generating a model of an object based on the training image; generating a modified training image based on the training image; determining a set of poses that represent possible instances of the model in the modified training image; and computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image. ### Claim 2 (depends on 1) The method of claim 1 further comprising computing at least a primary score and a secondary score for at least a portion of the set of poses. ### Claim 3 (depends on 2) The method of claim 2 wherein computing the quality metric is based on the primary score and the secondary score. ### Claim 4 (depends on 2) The method of claim 2 wherein computing the quality metric is based on a distribution of the computed scores for the portion of the set of poses. ### Claim 5 (depends on 1) The method of claim 1 wherein the model comprises a geometric description of the object in the training image. ### Claim 6 (depends on 1) The method of claim 1 wherein the model comprises a portion of the training image. ### Claim 7 (depends on 1) The method of claim 1 wherein generating the modified training image comprises adding noise to the training image. ### Claim 8 (depends on 7) The method of claim 7 wherein the noise comprises amplifier noise, salt-and-pepper noise, shot noise, quantization noise, film grain noise, non-isotropic noise, or any combination thereof. ### Claim 9 (depends on 7) The method of claim 7 wherein the noise is added to one or more pixels in the training image. ### Claim 10 (depends on 1) The method of claim 1 wherein generating the modified training image comprises transforming the training image by one or more degrees-of-freedom of rotation, translation, scale, skew, aspect ratio, or any combination thereof. ### Claim 11 (depends on 1) The method of claim 1 wherein generating the modified training image comprises changing the resolution of the training image. ### Claim 12 (depends on 1) The method of claim 1 further comprising: generating a plurality of modified training images based on the training image; and determining a set of poses for each of the plurality of modified training images, each set of poses representing possible instances of the model in one of the plurality of modified training images; wherein computing the quality metric is further based on an evaluation of the sets of poses, determined from the plurality of modified training images, with respect to the modified training image. ### Claim 13 (depends on 12) The method of claim 12 further comprising computing at least primary scores and secondary scores for at least a portion of the set of poses and at least portions of each of the sets of poses computed from the plurality of modified training images. ### Claim 14 (depends on 13) The method of claim 13 wherein computing the quality metric of the model is based on the secondary scores. ### Claim 15 (depends on 13) The method of claim 13 wherein computing the quality metric of the model is based on a distribution of the computed scores for the portion of the set of poses and distributions of the computed scores for the portions of the sets of poses computed from the plurality of training images. ### Claim 16 (depends on 1) The method of claim 1 further comprises modifying a baseline model parameter, wherein generating the model is based on the modified baseline model parameter. ### Claim 17 (depends on 16) The method of claim 16 wherein the baseline model parameter comprises an elasticity parameter, a grain limit or, granularity, a coarse-value acceptance fraction, a contrast threshold, an edge-value threshold, trainClientFromPattern or any combination thereof. ### Claim 18 A computer program product, tangibly embodied in a non-transitory machine-readable storage device, the computer program product including instructions being operable to cause a data processing apparatus to: receive a training image; generate a model of an object based on the training image; generate a modified training image based on the training image; determine a set of poses that represent possible instances of the model in the modified training image; and compute a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image. ### Claim 19 A system for determining a quality metric of a model of an object in a machine vision application, the system comprising: interface means for receiving a training image; model generating means for generating a model of an object based on the training image; image modifying means for generating a modified training image based on the training, image; processor means for determining a set of poses that represent possible instances of the model in the modified training image; and processor means for computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image. ### Claim 20 A machine vision system for determining a quality metric of a model of an object in a machine vision application, the system comprising: an interface for receiving a training image; a model generating module for generating a model of an object based on the training image; an image processing module for generating a modified training image based on the training image; a run-time module for determining a set of poses that represent possible instances of the model in the modified training image; and a quality-metric module for computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image.