Research Of Image Fusion Based On Salient Computation And Partial Differential Equation

As a branch of Information Fusion, the Image fusion technology aims to combine separate information obtained from multiple sensors into a synthetic image. This technique can preserve salient-object information of a source image as much as possible for further object-oriented detection, analyses or processing. It is widely used in computer, remote sensing, military, medical and other applications. O ne of the most popular topics in the Image Fusion research area is Multi-scale Image Fusion. However existing multi-scale fusion method does not include the brightness and salient-object differences of different images as a factor in the weighted low- frequency coefficient process. This result in decreased image contrast compares to the source image. This research introduce a novel method that generates more accurate weight for multi-scale transform, base on the partial differential equation(PDE) and total variation(TV) method. The Salient Computation is utilized at the same time to guide the fusion of low frequency coefficient and optimize the weights to further improve the fusion results.The major works carried out in this research are listed as below:(1) Analyze the multi-scale image fusion methods. Survey and summarizes the PDE and TV applications in the field of image processing. Introduce the visual salient computation method.This lay the foundation for solving practical problems in the image fusion process.(2) Analyze and summarize the modeling methods based on the TV and PDE theory.Traditional fusion process does not include brightness difference, the PDE and TV theory are adapted to build a new mathematical model to obtain ideal weight of low frequency coefficients in transform domain and improve fusion results.(3) Generates Saliency Maps utilizes salient computation, the improved result reflects the salient object of the source images. In addition to this, salient computation method is used to guide the fusion of low frequency coefficients so that salient-object information of source images are better protected.Image Fusion experiments are carried out utilizing visible light images, medical images and remote sensing images that is registration processed. Result shows that the proposed method can overcome the low image contrast problem, and the reconstructed image protects salient-object information,, such as shape, size, color, texture, etc effectively, and it is more effective for processing source images with brightness differences. The subjective visual perception and objective evaluation ind ex get certain improvement.