Study On Dehighlighting Technique Of Single Image

Image processing is becoming more and more important in everyday life and its effectiveness is closely related to the quality of the image,which is often unsatisfactory.According to the two-colour reflection model,the light on the surface of an object can be divided into specular and diffuse light.The diffuse light represents the colour information of the object itself,while the specular light is the highlight,and the intensity of the highlight component is much greater than that of the diffuse light.Therefore,when specular reflection occurs on the surface of an object,the textural and colour information of the object becomes weaker,to the extent that when the specular reflected light reaches saturation,the information on the surface of the object is completely lost.Therefore,it is particularly important to remove the highlighting component of the image and restore the original information to the image to improve the image quality.This paper investigates the highlight removal algorithm for single images,which mainly consists of highlight region detection and highlight region removal.The main algorithms used in highlight detection are the visual saliency detection algorithm(LC algorithm)and the region growing algorithm.the LC algorithm obtains the saliency value of an image by calculating the difference between the gray value of each pixel point in the highlight image and all other pixel points except itself and then summing them.in order to distinguish the highlight areas in the highlight image from other colour highlighted areas,before using the LC algorithm In order to distinguish highlight areas from other colour highlight areas,the LC algorithm first converts the original highlight image from RGB colour space to YCr Cb colour space,and then uses the salient values that meet certain conditions as the seed points for the region growth algorithm,which detects the highlight areas in the highlight image according to the selected initial seed points and the set growth criteria,but since the detected highlight areas have problems such as depressions and corner points,the diffuse water filling algorithm-the same areas are filled according to the difference in the gray value of the pixel values and,when using this algorithm,the filled areas are expanded appropriately to give a smoother detection image.The detection image is then used as the mask image for the highlight removal algorithm,the Criminisi algorithm.Since the effect of highlight removal obtained by the traditional Criminisi algorithm is not ideal,this paper improves it by introducing the structure tensor,mainly in the calculation of the priority,the matching criterion and updating confidence in three aspects.An experimental comparison of the proposed de-highlighting algorithm with other image de-highlighting algorithms confirms that the de-highlighting effect of this paper is better,resulting in a clearer image after removal.