发明名称 IMAGE RECOGNITION METHOD, IMAGE RECOGNITION DEVICE AND IMAGE RECOGNITION PROGRAM
摘要 The image recognition method includes: setting ROIs in an image of one image plane; and searching for a region including a recognition-targeted image. The method involves dictionary data including feature quantities calculated from blocks forming a recognition-targeted region including a recognition-targeted image and having blocks concerning the region. Focusing on one block in the image plane, feature quantities are calculated, which make block-of-interest feature quantities. With each of ROIs including the block, inner products of feature quantities calculated from a block corresponding to the one block in the dictionary data, and block-of-interest feature quantities are accumulatively added into evaluation value intermediate data for ROI. Whether or not each ROI includes the recognition-targeted image is determined on an evaluation value of ROI on which the accumulative addition has been finished on all of blocks forming the ROI. The noticed blocks are sequentially scanned in the image plane.
申请公布号 US2016283801(A1) 申请公布日期 2016.09.29
申请号 US201615064472 申请日期 2016.03.08
申请人 RENESAS ELECTRONICS CORPORATION 发明人 OKUMURA Shunsuke
分类号 G06K9/00;G06T1/60;G06K9/46;G06T7/00;G06K9/52;G06T7/60 主分类号 G06K9/00
代理机构 代理人
主权项 1. An image recognition device for executing an image recognition process, comprising: a processor; and a memory, wherein a program run by the processor executes the image recognition process, the image recognition process includes: setting regions of interest in an image of one image plane;searching for a region including a recognition-targeted image;defining blocks each composed of pixels; andholding, in the memory, dictionary data including feature quantities respectively calculated from blocks forming a recognition-targeted region corresponding to the recognition-targeted region including the recognition-targeted image and composed of blocks, in the image recognition process, each region of interest is composed of blocks of which the number is the same as the number of blocks forming the recognition-targeted region, the processor calculates feature quantities for each block in the image plane, which make block-of-interest feature quantities, the processor accumulatively adds, for each of regions of interest including a block concerned, a score calculated from feature quantities calculated from a block corresponding to the block concerned in the dictionary data, and the block-of-interest feature quantities into intermediate data of an evaluation value for the region of interest, temporarily holds the intermediate data in the memory, andmakes determination about whether or not each region of interest includes the recognition-targeted image based on an evaluation value of the region of interest on which the accumulative addition has been finished on all of blocks forming the region of interest.
地址 Tokyo JP