Image processing method and device based on fully convolutional neural network

Abstract

The present invention provides an image processing method and device based on a fully convolutional neural network. The method includes: performing data thinning on the image data of the image to be processed, and converting the image to be processed after data thinning into a single image corresponding to the three primary colors. Channel the grayscale image and perform image convolution to obtain the feature map of the image to be processed; perform image deconvolution on the feature map of the image to be processed to restore the feature map to the same resolution as the image to be processed and mark the category of the target object out the target image. Data thinning makes the pixel value of each pixel in the image more accurate, and more detailed features can be extracted when extracting features through convolution. Converting the three-channel image into three single-channel grayscale images corresponding to the three primary colors for convolution can reduce the loss of image features compared with direct convolution of the three-channel image, so that the final image processing result is more accurate and can be easily Good for handling images in low light conditions.

Description

technical field [0001] The present invention relates to the field of image processing, in particular to an image processing method and device based on a fully convolutional neural network. Background technique [0002] Image semantic segmentation is to identify the content of the image through automatic machine segmentation. It can be said to be the cornerstone technology of image understanding, and it is a pivotal application in the fields of automatic driving systems, drone applications, and wearable devices. [0003] Deep learning techniques can improve the performance of image semantic segmentation tasks, but most deep learning techniques (such as fully convolutional neural networks (fully CNN)) have good results on datasets of high-resolution images or depth images, but The effect of using deep learning technology for image semantic segmentation is not good when dealing with low-light, low-resolution or unevenly illuminated images. Contents of the invention [0004] ...

AI Technical Summary

Problems solved by technology

[0003] Deep learning techniques can improve the performance of image semantic segmentation tasks, but most deep learning techniques (such as fully convolutional neural networks (fully CNN)) have good results on datasets of high-resolution images or depth images, but When dealing with low-light, low-definition or unevenly illuminated images, the effect of using deep learning technology for image semantic segmentation is not good.

Images