# US 10,769,478 B2 · Convolutional neutral network identification efficiency increasing method and related convolutional neutral network identification efficiency increasing device > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,769,478 B2 - Document ID: US10769478B2 - Status: Granted - Kind: B2 - Application number: 16/197,361 - Filed: 2018-11-21 - Published: 2020-05-21 - Granted: 2020-09-08 - Earliest claimed: 2018-11-21 - Assignees: Vivotek Inc. - Applicants: Vivotek Inc. - Inventors: Cheng-Chieh Liu; Chia-Po Wei; Fu-Min Wang; Chia-Wei Chi - Web page: https://patentgrep.com/patent/US10769478B2 ## CPC - G06T 7/194: involving foreground-background segmentation - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 5/046: Forward inferencing; Production systems - G06V 10/28: Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns - G06V 10/507: {Summing image-intensity values; Histogram projection analysis} - G06V 10/764: using classification, e.g. of video objects - G06V 10/82: using neural networks - G06V 20/58: Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads - G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks - G06T 2207/10016: Video; Image sequence - G06T 2207/20084: Artificial neural networks [ANN] - G06T 2207/30196: Human being; Person - G06T 2207/30232: Surveillance - G06T 2207/30236: Traffic on road, railway or crossing - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract A convolutional neutral network identification efficiency increasing method is applied to a related device. The convolutional neutral network identification efficiency increasing method includes analyzing an input image to acquire foreground information, utilizing the foreground information to generate a foreground mask, and transforming the input image into an output image via the foreground mask. The output image is used to be an input of the convolutional neutral network identification for preferred object identification efficiency. ## Claims ### Claim 1 A convolutional neutral network identification efficiency increasing method, comprising: analyzing an input image to acquire foreground information; computing a histogram of the foreground information; dividing the histogram into at least one first group and at least one second group via its pixel value ranges, wherein a pixel value range of the first group is smaller than a pixel value range of the second group; comparing a pixel amount of the second group with a predetermined parameter; and generating a foreground mask according to a comparison result; and transforming the input image into an output image via the foreground mask, wherein the output image is used to be an input of convolutional neutral network identification for increasing object identification efficiency; wherein generating the foreground mask comprises: setting a foreground threshold when the pixel amount of the second group is greater than the predetermined parameter, defining pixels of the foreground information having pixel values greater than the foreground threshold as a first set of pixels, defining pixels of the foreground information having pixel values smaller than the foreground threshold as a second set of pixels, and setting pixels of the foreground mask which corresponds to the first set of pixels having pixel values as a first numeral, and setting pixels of the foreground mask which corresponds to the second set of pixels having pixel values as a second numeral; wherein generating the foreground mask further comprises: determining whether the first group conforms to a specific condition when the pixel amount of the second group is smaller than the predetermined parameter, and setting pixels values of all pixels inside the foreground mask as a first numeral when the first group conforms to the specific condition. ### Claim 2 (depends on 1) The convolutional neutral network identification efficiency increasing method of claim 1, further comprising: utilizing object detection technique to set a range of the input image within a monitoring frame. ### Claim 3 (depends on 1) The convolutional neutral network identification efficiency increasing method of claim 1, wherein analyzing the input image to acquire the foreground information comprises: generating background information of the input image; and computing a difference between the input image and the background information to acquire the foreground information. ### Claim 4 (depends on 1) The convolutional neutral network identification efficiency increasing method of claim 1, wherein pixels values of all pixels inside the foreground mask are set as a second numeral when the first group does not conform to the specific condition. ### Claim 5 (depends on 1) The convolutional neutral network identification efficiency increasing method of claim 1, wherein transforming the input image into the output image via the foreground mask comprises: computing products of all pixel values inside the input image and corresponding pixel values inside the foreground mask respectively; defining some of the products which have positions corresponding to pixel positions inside the foreground mask not belonging to a second numeral as a first set of products; defining some of the products which have positions corresponding to pixel positions inside the foreground mask belonging to the second numeral as a second set of products; replacing the second set of products with a reference value; and utilizing the first set of products and the reference value to generate the output image. ### Claim 6 (depends on 1) The convolutional neutral network identification efficiency increasing method of claim 1, wherein a grey level of pixels inside the output image is ranged from 0 to 128. ### Claim 7 (depends on 1) The convolutional neutral network identification efficiency increasing method of claim 1, wherein a dimension of the foreground mask is identical with a dimension of the input image. ### Claim 8 A convolutional neutral network identification efficiency increasing device, comprising: an image generator adapted to acquire an input image; and an operational processor electrically connected to the image generator, and adapted to analyze an input image for acquiring foreground information, compute a histogram of the foreground information, divide the histogram into at least one first group and at least one second group via its pixel value ranges, compare a pixel amount of the second group with a predetermined parameter, and generate a foreground mask according to a comparison result, and transform the input image into an output image via the foreground mask, wherein the output image is used to be an input of convolutional neutral network identification for increasing object identification efficiency; wherein a pixel value range of the first group is smaller than a pixel value range of the second group; wherein the operational processor is further adapted to set a foreground threshold when the pixel amount of the second group is greater than the predetermined parameter, define pixels of the foreground information having pixel values greater than the foreground threshold as a first set of pixels, define pixels of the foreground information having pixel values smaller than the foreground threshold as a second set of pixels, and set pixels of the foreground mask which corresponds to the first set of pixels having pixel values as a first numeral and set pixels of the foreground mask which corresponds to the second set of pixels having pixel values as a second numeral; wherein the operational processor is further adapted to determine whether the first group conforms to a specific condition when the pixel amount of the second group is smaller than the predetermined parameter, and set pixels values of all pixels inside the foreground mask as a first numeral when the first group conforms to the specific condition. ### Claim 9 (depends on 8) The convolutional neutral network identification efficiency increasing device of claim 8, wherein the operational processor is further adapted to generate background information of the input image, and compute a difference between the input image and the background information for acquiring the foreground information. ### Claim 10 (depends on 8) The convolutional neutral network identification efficiency increasing device of claim 8, wherein pixels values of all pixels inside the foreground mask are set as a second numeral when the first group does not conform to the specific condition. ### Claim 11 (depends on 8) The convolutional neutral network identification efficiency increasing device of claim 8, wherein the operational processor is further adapted to compute products of all pixel values inside the input image and corresponding pixel values inside the foreground mask respectively to generate the output image. ### Claim 12 (depends on 8) The convolutional neutral network identification efficiency increasing device of claim 8, wherein the operational processor is further adapted to compute products of all pixel values inside the input image and corresponding pixel values inside the foreground mask respectively, define some of the products which have positions corresponding to pixel positions inside the foreground mask not belonging to a second numeral as a first set of products, define some of the products which have positions corresponding to pixel positions inside the foreground mask belonging to the second numeral as a second set of products, replace the second set of products with a reference value, and utilize the first set of products and the reference value to generate the output image. ### Claim 13 (depends on 8) The convolutional neutral network identification efficiency increasing device of claim 8, wherein a grey level of pixels inside the output image is ranged from 0 to 128. ### Claim 14 (depends on 8) The convolutional neutral network identification efficiency increasing device of claim 8, wherein a dimension of the foreground mask is identical with a dimension of the input image. ### Claim 15 A convolutional neutral network identification efficiency increasing method, comprising: analyzing an input image to acquire foreground information; computing a histogram of the foreground information; dividing the histogram into at least one first group and at least one second group via its pixel value ranges, wherein a pixel value range of the first group is smaller than a pixel value range of the second group; comparing a pixel amount of the second group with a predetermined parameter; and generating a foreground mask according to a comparison result; and computing products of all pixel values inside the input image and corresponding pixel values inside the foreground mask respectively to generate an output image, wherein the output image is used to be an input of convolutional neutral network identification for increasing object identification efficiency. ### Claim 16 (depends on 15) The convolutional neutral network identification efficiency increasing method of claim 15, further comprising: utilizing object detection technique to set a range of the input image within a monitoring frame. ### Claim 17 (depends on 15) The convolutional neutral network identification efficiency increasing method of claim 15, wherein analyzing the input image to acquire the foreground information comprises: generating background information of the input image; and computing a difference between the input image and the background information to acquire the foreground information. ### Claim 18 (depends on 15) The convolutional neutral network identification efficiency increasing method of claim 15, wherein transforming the input image into the output image via the foreground mask comprises: computing products of all pixel values inside the input image and corresponding pixel values inside the foreground mask respectively; defining some of the products which have positions corresponding to pixel positions inside the foreground mask not belonging to a second numeral as a first set of products; defining some of the products which have positions corresponding to pixel positions inside the foreground mask belonging to the second numeral as a second set of products; replacing the second set of products with a reference value; and utilizing the first set of products and the reference value to generate the output image. ### Claim 19 (depends on 15) The convolutional neutral network identification efficiency increasing method of claim 15, wherein a grey level of pixels inside the output image is ranged from 0 to 128. ### Claim 20 (depends on 15) The convolutional neutral network identification efficiency increasing method of claim 15, wherein a dimension of the foreground mask is identical with a dimension of the input image.