Backlight Image Enhancement By Combining Multi-image Fusion With Guided Filtering

With the continuous progress of the times,the hardware of the shooting tools is changing and the images are getting clearer and clearer.People are becoming more and more demanding for the quality of the image.The backlight image is produced because the shooting environment leads to the dark brightness of the subject area and the serious loss of the color and details,while the brightness of the background area is large and the color is oversaturated,which greatly reduces the image quality and can not meet the requirements of the people.Therefore,the research of backlight image enhancement algorithm is of great significance.At present,there are few researches on Algorithms for backlight image enhancement.The backlight image is similar to the uneven illumination image,and the image enhancement is processed by using an uneven illumination image enhancement algorithm,such as image enhancement based on histogram equalization.However,the results of this kind of algorithm will improve the overall brightness of the image,and have better processing results for the backlight image with a larger proportion in the less exposed area,but it is not ideal for the backlight image processing which has a large proportion of over exposure area,which will lead to the reduction of the details of the overexposure area.The backlight image can also be considered as a combination of low illumination and overexposure images.The image is processed with a low illumination image enhancement algorithm and an overexposure image enhancement algorithm.The result of such processing can make the whole image discontinuous,although the details of the under exposed and overexposed areas are enhanced.In view of the above problems,the following studies are carried out in this paper:(1)In order to enhance the backlight image intelligently,this paper proposes a backlight image detection algorithm based on SVM combined with the details and spatial features of the backlight image.In this paper,two backlight parameters are used:backlight parameters based on detail and backlight parameters based on spatial characteristics.The two parameters of the normal illumination and backlight images are sent to SVM to train the segmentation line to distinguish the backlight image and the normal illumination image.The accuracy of the method has been greatly improved.The two backlight parameters of the multiple images are sent into the SVM to be trained to get the segmentation line of the backlight image and the normal illumination image,in order to detect whether the image is a backlight image.The use of SVM for backlight image detection has achieved high accuracy.(2)The common image enhancement algorithms are used to deal with the backlight image,for example,the image enhancement based on adaptive histogram equalization,the image enhancement based on the discrete wavelet image fusion and the MSRCR based image enhancement.The advantages and disadvantages of each algorithm in the process of backlight image processing are analyzed.(3)In view of the shortcomings of the existing image enhancement algorithm,a new image enhancement algorithm combining multi exposure image fusion and guided filtering is proposed in this paper.In this method,a number of pseudo exposure images are generated by a single image,a pseudo exposure image is processed with a guide filter,and then the image is partitioned.In the process of segmentation,an image fusion algorithm based on the subblock variance is proposed for the image fusion to generate the enhanced image.By comparing the other algorithms,it shows the effectiveness of the algorithm.(4)Using MATLAB software,a backlight image enhancement system is developed and the evaluation criteria are given.At present,the objective evaluation of the quality of an image is luminance mean,standard deviation,information entropy,PSNR,clarity and so on.This paper evaluates the enhancement effect by means of luminance mean and JND critical deviation.