Research On Car Body Color Recognition Algorithms Based On Defogging And Highlight Inpainting

Since the rapid development of intelligent transportation system, the car feature recognition has got more and more attention. The related recognition techniques that include license plates, car body color, car models and so on are being undertaken to reinforce the intelligent traffic management. Among the related techniques the algorithm of license plate recognition has been relatively well developed, however only relying on license plate recognition cannot meet today’s vehicle management. The images captured under foggy and highlight environment will appear color distortion, so the real car body color cannot be recognized. Considering all these cases,this paper proposes a car color recognition algorithm based on defogging and highlight inpainting to assist license plate recognition to manage vehicles, which can be used at highway toll gate and parking entrance to gather and extract the color information of the vehicle and recognize the car body color accurately in complex environment.Firstly, the overall flowchart of the car body color recognition algorithms based on defogging and highlight inpainting is described, including image preprocessing and several typical color spaces. Secondly, the research on the car body recognition of atomized images is proposed. The paper presents a defogging method of atmospheric dissipation model on the base of dark channel prior theory combining atmospheric light scattering model and atmospheric dissipation function to recover the image from atomized image. Additionally,weight block and HSV color space histogram are adopted to recognize the body color to avoid the influence of background colors. Experimental results show that after defogging process the body color can be recognized fast and accurately and the algorithm not only can reduce the background interference but also can increase the accuracy of color recognition. Finally,the paper proposes a highlight inpainting algorithm based on improved Criminisi theory and car contour aiming at the images affected by highlight. The algorithm marks the areas affected by highlight using saliency detection method in YUV color space and improved Criminisi algorithm and car contours are adopted to restore the car highlight area and inpainte the color information of the car. After inpainting of highlight area, weight block and HSV color space histogram are adopted to recognize the car again. Experimental results show that highlight inpainting algorithm of the car contributes to the recognition of car images affected by highlight and the recognition accuracy is increased by 8.7%.The research on image processing and color recognition is proposed in this paper aiming at the car images collected in complex environment. A car color recognition algorithm based on defogging and highlight inpainting is presented to avoid the influences of foggy and highlight environment. Where weight block is adopted to decrease the interference of background color and defogging algorithm based on atmospheric dissipation model is presented to restore the color information from atomized image. In addition, highlight area is marked using highlight detection method of YUV color space and improved Criminisi algorithm is proposed combining car contour to remove the influence on car body color recognition. The proposed car body color recognition algorithm can achieve high color recognition accuracy and has the strong robustness with the practical background significance.