A Color Image Semantic Classification Method Based on Fully Convolutional Network

Abstract

The invention discloses a color image color semantic classification method based on a full convolution network, comprising: 1 constructing a full convolution network; 2 obtaining a color image data set with pixel-level annotation; 3 using the color image data set to fully The network is trained to obtain a feature model that can perform pixel-level color semantic classification for color images of any size; 4. Use the feature model to perform pixel-level color semantic classification for any color image, and evaluate the classification accuracy of the feature model; 5. Use the full connection condition to randomly The airport method optimizes the network classification results to obtain the color category label of each pixel in the image, and converts the color category label to the corresponding color space according to the mapping relationship between the category label and the color space to display pixel-level color semantic classification result. The invention can realize the color semantic classification of the color image pixel level, and effectively improve the accuracy of the color semantic classification of the color image in complex and changeable environments.

Description

technical field [0001] The invention belongs to the field of computer / machine vision, image processing and analysis, in particular to a method for color semantic classification of color images based on a fully convolutional network. Background technique [0002] In computer vision, color is an important attribute of images and an important way for humans to perceive image information. By assigning image color category labels, it can be further applied to image retrieval, image annotation, color blindness assistance, visual tracking, language human-computer interaction and other fields. Therefore, good color semantic classification results are helpful for further image processing and image analysis. [0003] Existing image color semantic classification methods include methods based on statistical models and methods based on deep learning. [0004] Methods based on statistical models are mainly based on color stimuli, such as color semantic classification by perceiving three...

AI Technical Summary

Problems solved by technology

[0009] In order to solve the deficiencies in the prior art, the present invention provides a color image color semantic classification method based on a fully convolutional network, aiming at solving the color image pixel-level color attribute semantic classification problem, by constructing a fully convolutional neural network , to obtain a color attribute semantic classification network feature model with good classification accuracy in complex scenes, thereby improving the accuracy of color semantic classification of color images in complex and changeable environments

Images