Research On Image Fusion Algorithms

Image fusion is to synthesize multiple source images with complementary information representation to a new image and get a clear and complete knowing than a single image. In recent years, image fusion has become a new important and useful technology in the field of image processing and computer vision. This thesis has introduced the principle of the pulse coupled neural networks(PCNN), Contourlet, and nonsubsampled Contourlet. Some research based on these theories in pixel-level and feature-level fusion has conducted. The primary contributions of this paper contain the following points:1. Pulse coupled neural networks is a visual cortex-inspired neural network and characterized by the global coupling and pulse synchronization of neurons. It has been successfully employed in image processing. In view of the characteristic of the infrared and visible light images, an image fusion strategy which uses PCNN to segment goal from the infrared image, and uses structural similarity property to eliminate the false goal is proposed. The background of the fusion image directly comes from the visible light image.2. The relations of the iterative times and wavelet decomposition layers is analyzed in this paper, as well as the relations between the frequency band and the fire times. A new measure is proposed when the traditional fusion rule has not considered the same and near fire times in the peculiar circumstance.3. The phase information and local relativity is considered in the region energy fusion rule of the nonsubsampled Contourlet transform. An improved weighted average fusion rule is proposed to avoid the limitation of detail blurring effect while increasing the entropy of the fusion image. Consistency check strategy is used in fusion coefficient to avoid discontinuity of the coefficient.