Feature similarity-based non-local neighborhood gray level image colorization method

Abstract

The invention provides a feature similarity-based non-local neighborhood gray level image colorization method. A gray level image is given by a user as an input image; a marked feature direction field is obtained through the input image; a textural feature vector field is obtained according to the marked feature direction field and the input image; a structure chart is obtained by calculating the input image with a relative total variation method; a high-dimensionality feature vector field is constructed through the structure chart, the textural feature vector field, the gray level image and an image space coordinate system; the nearest neighbor of each pixel is obtained through the feature vector field; the user performs little artificial coloring on the input image to obtain a colored image, and converts the input image and the colored image to a Lab color space; in combination of the colored image and the nearest neighbor of each pixel, a sparse linear equation set is constructed, and the sparse linear equation set is solved to obtain a target image; and the target image is converted to an RGB color space to obtained a colorized image. The method takes brightness, coordinates and textual features into account, and improves a colorization effect.

Description

technical field [0001] The invention relates to a grayscale image colorization method, in particular to a non-local neighborhood grayscale image colorization method based on feature similarity. Background technique [0002] Colorization is a computer-aided process that adds color to a black and white image or video. This technology is currently widely used in the fields of film and television, medical treatment, space exploration, industry, and science. Generally, colorization methods are divided into two categories: colorization methods based on reference pictures and colorization methods based on user graffiti interaction. The colorization method based on the reference picture is the process of transferring the color in the reference picture to the target image according to a certain degree of correlation. Although this type of method can obtain better colorization results, it relies heavily on reference pictures, thus restricting the development of this method. However...

AI Technical Summary

Problems solved by technology

This method has two main disadvantages: on the one hand, because the method is based on a local colorization algorithm, the colorization results of each area are still relatively dependent on user input, and color bleeding is prone to occur

However, this method also does not take into account the texture information of the image

At the same time, when the areas with the same brightness of the image are painted with different color lines, the colorization result will appear color mixture

Images