Research On Colorization Method Of Polarization Image Based On Color Migration

The polarization characteristics of light determine that polarization imaging technology has the unique advantage of describing and analyzing image information.However,most polarization images are presented in gray images.How to make full use of the color recognition ability of human vision,interpret the polarization grayscale information,and realize the image colorization of the polarization image is a new proposition in the field of polarization image processing.To highlight the advantages of polarization imaging,obtain color polarization image suitable for detection and recognition by the human eyes.In this paper,a research on the colorization method of polarization image based on color migration is carried out,the information fusion of light intensity image and polarization degree image is carried out,the obtained polarization feature fusion image is used as the processing object in the colorization task.The main research contents of this paper are as follows:1)Aiming at the problems that the traditional polarization image coloring method only performs global mapping and does not consider local features,resulting in more distortion of the pseudo color image and mistransmission of colors,a color migration method for polarization image based on graph-theoretic segmentation and gray level co-occurrence matrix is proposed.Firstly,the target polarization image and the reference image are divided into several sub-regions by using the graph-theoretic segmentation method in the YCb Cr color space,and then by calculating the texture feature statistics of the gray level co-occurrence matrix,the similar sub-regions of the two images are matched pair by pair.Finally,color migration is performed.The experimental results show that the method in this paper can effectively alleviate the problem of color mistransmission,the color of the polarization image is in line with human visual perception,and the image texture details are clear.2)Aiming at the mistransmission of color in traditional color migration methods and the insufficient understanding of polarization image features by deep learning color migration methods,resulting in insufficient similarity feature extraction and color overflow in the resulting image,a polarization image color migration method based on dual-branch convolutional neural network is proposed.The similarity feature extraction network at the front end of the network has a dual-branch structure,and the two branches are: SE-Res Net local feature extraction network with "feature recalibration" and improved VGG-19 global feature extraction network with classification properties.The local features,global features,similarity bidirectional mapping function,and aligned reference chromaticity are fused,and the fusion result is used as the input of the colorization network.Finally,the colorization network combines the polarization features to perform color migration on the target polarization image.The chroma loss function and perceptual loss function at training time guarantee chroma consistency and robustness of color migration.The experimental results show that the method in this paper can effectively solve the problem of color overflow and the color distribution is natural.Compared with the visible light image colorization results,the polarization image colorization results retain more details and texture information,which is more conducive to the detection and recognition of human vision.