Image recognition device and image recognition program

Patent metadata

Field	Value
Publication number	US-2022189134-A1
Application number	US-202017599323-A
Country	US
Kind code	A1
Filing date	Mar 30, 2020
Priority date	Mar 28, 2019
Publication date	Jun 16, 2022
Grant date	—

Abstract

An image recognition device involves successively extracting co-occurrence pairs in synchronization with a clock, setting a weighting for the portion connecting the input layer and the intermediate layer corresponding to the extracted co-occurrence pairs, and successively inputting a first vote to the input layer. Meanwhile, the intermediate layer adds and stores the successively inputted number of votes. By continuing this operation, a value the same as if a histogram were inputted to an input layer is achieved in the intermediate layer, without creating a histogram. In this way, the image recognition device of this embodiment can perform image recognition while avoiding the creation of a histogram, which consumes vast amounts of memory. As a result of this configuration, it is possible to save memory resources, simplify circuits, and improve calculation speed, and achieve an integrated circuit suitable to an image recognition device.

First claim

1 . An image recognition device comprising:
feature element acquiring means for acquiring feature elements of an image recognition target; and image recognition means which has an addition layer to which the feature elements are directly input and which weights a plurality of values to the feature elements, and which recognizes an identification target with the use of an addition value obtained by sequentially adding the feature elements in the addition layer.

2 . The image recognition device according to claim 1 , wherein the image recognizing means is a binary network which assigns binary weights to the feature elements.

3 . The image recognition device according to claim 1 , wherein the image recognizing means comprises image recognition result outputting means for acquiring a total value of addition values obtained by sequentially adding the feature elements, and then fixing an image recognition result of the image recognition.

4 . The image recognition device according to claim 1 , wherein, the image recognizing means sequentially updates the image recognition result corresponding to the total value of the addition value obtained by sequentially adding the feature elements in connection with the update of the storage, and outputs the image recognition result if the updated image recognition result is the same for a predetermined number of consecutive times.

5 . The image recognition device according to claim 1 , comprising:
image acquiring means for acquiring an image; and pixel value acquiring means for acquiring luminance as a pixel value of a pixel in the acquired image, wherein the feature element acquiring means sequentially acquires a plurality of types of feature elements of feature amounts representing features of the image recognition target in accordance with each feature element based on a luminance gradient direction using the acquired luminance, and the image recognizing means performs sequential additions in the addition layer in accordance with each feature element.

6 . The image recognition device according to claim 5 , wherein the feature element acquiring means acquires the feature elements based on co-occurrences of the luminance gradient directions.

7 . The image recognition device according to claim 6 , wherein the image acquiring means acquires images with different resolutions of the same subject, and
the feature element acquiring means acquires the feature elements based on co-occurrences of the images with the different resolutions of the luminance gradient directions.

8 . The image recognition device according to claim 1 , comprising:
selecting means for selecting a predetermined feature element of the acquired feature elements and inputting it to the addition layer.

9 . The image recognition device according to claim 1 , comprising:
replicating means for sequentially replicating the acquired feature elements and inputting them to the addition layer.

10 . An image recognition program for causing a computer to realize:
a feature element acquiring function which acquires feature elements of an image recognition target; and an image recognizing function which has an addition layer to which the feature elements are directly input and which weights a plurality of values to the feature elements, and which recognizes an identification target with the use of an addition value obtained by sequentially adding the feature elements in the addition layer.

Assignees

• Aisin Corp
• Kyushu Inst Tech

Inventors

• Yamada Hideo
• Muramatsu Ryuya
• Shibata Masatoshi
• TAMUKOH Hakaru

Classifications

• G06V10/955

  using specific electronic processors · CPC title

• G06V10/267 Primary

  by performing operations on regions, e.g. growing, shrinking or watersheds · CPC title

• G06V10/443

  by matching or filtering · CPC title

• G06V10/507 Primary

  Summing image-intensity values; Histogram projection analysis · CPC title

• G06T7/11

  Region-based segmentation · CPC title