Research On Pixel-Level Image Fusion

The research presented in this thesis is focused on the technique of pixel-level image fusion. Multi-sensor pixel-level image fusion is a promising technique. The aim of multi-sensor pixel-level image fusion is to integrate complementary and redundant information from multiple images to create a composite that contains a 'better' description of the scene than any of the individual source images. The composite image is suitable for observation or subsequent processing. Image fusion plays important roles in many different fields such as defense system, computer vision, remote sensing, concealed weapon detection and biomedical imaging, etc. In recent years, many image fusion systems have been developed. However, there are lots of problems need to be solved in the filed of multi-sensor pixel-level image fusion.In this thesis, in-depth, systematic and comprehensive research work has been done on multi-sensor image fusion, multi-focus image fusion based on multi-scale decomposition, autofocusing and artificial neural networks.The main contributions of this thesis are summarized as follows:1. The measure of image block clarity (or pixel clarity) in the multi-focus image fusion has been studied in this thesis. Several focus measures were studied in this paper as the measures of image clarity, in the field of multi-focus image fusion. All these focus measures are defined in the spatial domain and can be implemented in real-time fusion systems with fast response and robustness. This paper proposed a method to assess focus measures according to focus measures' capability of distinguishing focused image blocks from defocused image blocks.2. This thesis presents a method for multi-focus image fusion by using pulse coupled neural network (PCNN). The energy of image Laplacian is used as the measure of image clarity. The registered source images are first decomposed into blocks and the size of the image blocks is 8×8 pixels. Feature maps are obtained by computing the energy of image Laplacian of each block. Input the feature maps into PCNN as external stimulus. The final fused image can be constructed by selecting the image blocks from the source images based on the comparison of the outputs of the PCNN. Experimental results show that the proposed method outperforms some previous fusion methods, both in visual effect and objective evaluation criteria.3. The fusion of low-frequency band based on multi-scale decomposition has been studied in this thesis. We decomposed the image formation model, proposed by Sharma, into two parts, one part for the low-frequency band, the other part for the high-frequency band. The fusion of the low-frequency band is based on the low-frequency band image formation model. Experimental results show that the proposed method can preserve the image contrast.4. The fusion of high-frequency band based on multi-scale decomposition has been studied in this thesis. A probabilistic method is proposed for the fusion of high-frequency band based on the joint probability distribution of a wavelet coefficient pair—a wavelet coefficient and its parent. Experimental results show that the proposed method is robust to noise.5. An objective image fusion performance measure is proposed in this thesis. This performance measure is composed of two parts, one part is a kind of objective evaluation measure considering image edges, the other part measure the area of object.