Variation Method And Partial Differential Equation-based Image Enhancement And Fusion

With the development of science and technology, image processing has been applied to more and more fields. Early image processing methods based on linear theory are unable to meet the requirements of practical application, so a variety of nonlinear algorithm are proposed, including probability theory, wavelet theory,morphology, variation method and partial differential equation, et al. With mature mathematical theory foundation, various forms of model and the rapid numerical algorithm, the variation method and partial differential equation method are introduced to the field of image processing and computer vision, undoubtedly provides powerful mathematical tool for the field, having become a hot research topic in recent years. This thesis focuses on some key problems of image enhancement and fusion based on the variation method and partial differential equation, and the main work and innovation are as follows:In accordance with the requirements of practical application, image enhancement can highlight some specific information of the image, providing high quality image information for the following image processing. At present, variation method and partial differential equation method-based image detail enhancement mainly uses the feature of pixel to design adaptive diffusion function to retain image details or smooth noise. Human vision is sensitive to the changes of local image details, which are actually image gradients. And by adjusting image gradients, image details can be enhanced. For the infrared images with fuzzy edge details, low contrast and big noise,based on gradient field reconstruction framework, gradient field nonlinear transform for infrared image enhancement is proposed. By constructing a decreasing function(? 1) as the coefficients of the original image gradient field, weak useful details canbe enhanced. The histogram equalization is applied to image gradient histogram.Considering the gradient histogram equalization makes the gradient value too large and the image is over-enhanced and has noise. In order to overcome the above disadvantage, by setting adaptive dual thresholds to qualify the gradients, the image is prevented from over enhancement. The total variation(TV) model is adopted in the reconstruction of the enhanced image to suppress noise.Due to uneven illumination, image details are buried in the dark background or bright background. Gradient field specification-based image enhancement algorithm is proposed. The histogram specification method can guide the mapping of the gradient value through a predetermined function to make the histogram become a predetermined shape. If the function is selected properly, the important gradient-scale range can be highlighted to reach the purpose of enhancing images. By analyzing the characteristics of the gradient histogram, a Gaussian function is constructed to obtain the gradient histogram specification and therefore obtain the transform gradient field.In addition, sub-histogram equalization is proposed based on the histogram equalization to improve the image contrast, enhancing the details of dark or bright background. Further, double-peak Gauss function is constructed to fit the gradients of the targets and weak details in the gradient histogram. Then, enhance the two parts in different degrees, respectively. This can effectively enhance the weak details while avoiding over-enhancement of target edge details.Image fusion comprehensively uses the complementary information of different source images to synthesis a more complete and accurate image, aiming to make the information of the same scene can be more efficiently processed by computer and have better human perception. Image fusion technology has been widely used in military, medical and scientific research, et al. In recent years, researchers propose the image fusion methods in the gradient domain. Human vision is sensitive to the changes of image local details, which are actually image gradients. Image fusion in the gradient domain easily incorporate edge details of multiband images into a gradient field, making the reconstructed fused image retain the main features of the source images. Therefore, the key problem is to establish the fused gradient field.Variation multi-source image fusion based on the structure tensor algorithm is proposed, which can keep the image features and details very well. We first narrative the fusion gradient field based on structure tensor, and then measure characteristic graphs of each source image, constructing a weight value for the source image gradient according to the characteristic graph. Gradients with high image features are highlighted in the fusion gradient field, and thereby image features in the sources are well preserved. By using variation partial differential equation, the fusion image is reconstructed from the target gradient field. Based on actual experiment results, the average gradient value and entropy of the fused image are higher than the wavelet transform algorithm, tower decomposition algorithm and direct gradient fusion algorithm, and visual effect of the fusion image is good enough to retain features of source images and details. Therefore, it gives qualified image information for target detection and identification.For noisy image fusion, noise generally concentrates in high frequency part of the image, and easily be mistaken as the useful characterizations of the image, reducing the effect of image fusion significantly. Therefore, noisy image fusion has become challenging work in the field of computer vision. Gradient entropy metric and p-Laplace diffusion constraint-based algorithm is proposed for multispectral image fusion and de-noising. Firstly, fused contrast gradient field is constructed. To minimize the negative effects of noise on the selections of image features, the gradient entropy metric is proposed to construct the weight for each gradient of input images.Particularly, the local adaptive p-Laplace diffusion constraint is constructed to further suppress noise when rebuilding the fused image from the fused gradient field.Experimental results show that the proposed method effectively preserves edge detail features of multi-spectral images while suppressing noise.