A method for removing highlight of image in natural scene

Abstract

The invention discloses a method for removing highlight of an image in a natural scene, which comprises the following steps: decomposing an original image by using a robust sparse decomposition methodto obtain low-frequency information representing brightness change and color change of the image; converting the original image from a color space to a chrominance space, and estimating a specular reflection coefficient in the chrominance space by using low-frequency information; and calculating a specular reflection component by using the estimated specular reflection coefficient, removing the calculated specular reflection component from the original image, and converting the specular reflection component into a color space to obtain a highlight-removed image. According to the method, specular reflection can be accurately separated; the method has the advantages that the artifacts cannot be introduced during highlight elimination, the image cannot become blurred, the optimal highlight removal effect is achieved in a natural scene, and in addition, details and structural information of the original image are greatly reserved in some challenging scenes containing complex textures or saturated pixels.

Description

technical field [0001] The invention relates to the technical field of image quality restoration and improvement, in particular to a method for image highlight removal in natural scenes. Background technique [0002] During the shooting process of the image, the highlight phenomenon is often formed due to the reflection characteristics of the surface of the object in the scene and uneven lighting factors, that is, a relatively strong specular reflection occurs somewhere, which causes the original information of the image to be lost, thereby affecting the subsequent edge detection, Therefore, it is of great research significance to explore effective methods to accurately remove highlights in images or separate specular reflection and diffuse reflection components at pixels. [0003] There are two types of highlight removal methods for images, including hardware and software. The hardware method, such as adding a polarization device outside the camera, etc., the lack of such m...

AI Technical Summary

Problems solved by technology

[0004] The first method is based on partial differential equations and polarization, using polarization to separate reflection components, but the experimental process is more complicated, not only need to obtain multiple images with different polarization directions, but also need additional polarizer assistance, and this method only Suitable for situations where the surface of the object does not change significantly

[0005] The second method is based on the chromaticity diagram and histogram. The researchers proposed the concept of SF (no specular) image. By comparing the pixel values ​​​​in the highlight area and adjacent areas, the maximum value of the pixel is converted to match the adjacent pixel value. Constantly iterates to remove highlights, however the main disadvantage of this method is the blurring of hue saturation present in many natural images

The latest research method introduces color line constraints into the model, but it is only suitable for dark surface objects and cannot achieve accurate clustering of pixels

In addition, existing methods based on two-color reflectance models often assume that the highlights are sparse in the image, and convert the reflection separation into a low-rank and sparse matrix solution. However, these methods are often only effective for images taken in experimental scenarios. Because this is a scene where the environment is artificially arranged and light sources are added, however, the highlights in natural images usually do not show sparse characteristics, resulting in poor processing effects of existing methods in natural images

Images