Research On Night Vision Anti-halation Algorithm Based On Heterogeneous Image Fusion

When driving at night,the driver’s vision will be haloed by the strong light of the opposing vehicle,and it is impossible to see the road conditions ahead,which is very prone to traffic accidents.In order to solve the problem of halo in driving at night,this paper makes full use of the advantages of night vision infrared images that are not interfered by halo and visible light images have rich color information.By studying the anti-halation algorithm of heterogeneous image fusion,the goal of completely eliminating the blooming and improving driving safety at night is achieved.Although the existing image fusion anti-halation algorithm has achieved good results,it still has the following problems:1)The image enhancement algorithm enlarges the halo area in the image while enhancing the dark information of the visible light image,and the enhanced night vision infrared image has noise;2)The existing anti-halation fusion algorithm does not consider the saliency of the important information of the image,and the important features such as vehicles and pedestrians in the fused image are not prominent;3)There are big differences in the various quality evaluation indexes of the anti-halation fusion image,and no unified evaluation criteria are given,which affects the accuracy of the evaluation results.This article has carried out the following research work to address the above-mentioned problems:1)Aiming at the problem that existing image enhancement algorithms are not suitable for enhancing night vision halo scene images,by studying the gray difference between halo and non halo parts of night vision images,the halo critical gray level is constructed according to the degree of halo of the image value.Combining the halo critical gray value with the transmittance in the dark primary color prior enhancement algorithm,the transmittance that is automatically adjusted with the image halo intensity is designed on the basis of the dark primary color prior enhancement algorithm.Realize the self-adaptive enhancement of night vision halo image.2)In order to improve the importance of salient features such as vehicles and pedestrians in anti-halation fusion images,a YUV-FNSCT anti-halation fusion algorithm based on visual saliency is designed.By detecting the salient features of the infrared image,the FNSCT algorithm is used to fuse the infrared salient feature image and the visible light image brightness component decomposed by the YUV transformation to prevent halation.In the fusion process,a fusion strategy with automatic adjustment of low-frequency weights is designed,which eliminates halo and improves the color information of the fused image,thereby improving the human vision’s ability to recognize important information features in halo scenes.Thereby improving the human vision’s ability to recognize important information features in halo scenes,and achieving improved driving safety at night.3)In order to improve the accuracy of the anti-halation fusion image quality evaluation results,make full use of the advantages of the deep learning network that can automatically extract image features through convolution calculations.Aiming at the problem of the limited amount of anti-halation fusion image data,a convolutional neural network based on transfer learning is used to classify and evaluate the quality of the anti-halation fusion image.Through the learning and training of small samples of anti-blooming image data on the complex network structure,the goal of improving the accuracy of the anti-blooming fusion image quality evaluation results is achieved.The adaptive dark primary color a priori enhancement algorithm proposed in this paper improves the clarity and brightness of the night vision halo scene image,and avoids over-enhancing the halo area of the image.The proposed YUV-FNSCT anti-halation algorithm based on visual saliency can eliminate the fusion image blooming thoroughly and improve the significant information such as color details.The evaluation results obtained by the proposed anti-halation image quality evaluation model based on transfer learning are consistent with the subjective visual evaluation results of human eyes,which effectively improves the accuracy of the anti-halation fusion image quality evaluation results.