Image Mosaic And Fusion Technology Of Large Field View Multi-spectral Camera

Large field view multi-spectral camera is a new generation product. MS and PAN images of different pixel resolutions can be achieved. The camera focal plane is mosaicked by multi-chip TDICCD, and the multi-source detector is formed in parallel processing mode. Multi-source remote sensing images have different spatial resolution and spectral resolution, and the use of image fusion techniques can combine their respective strengths to make up for the lack of a single image information. Image mosaic and registration technologies will stitch together the overlapping areas of two images by removing redundant information. The following four questions are solved in this paper: 1) the registration of single channel MS and PAN images; 2) the registration and seamlessly mosaic of multi-channel images; 3) the fusion of multi-channel MS and PAN images; 4) the recoloring of fusion image based on color transfer.The mosaic and fusion technologies of multi-channel MS and PAN images from large field view camera are studied in this paper. The main contents of this paper include the following aspects:1) According to the structure characteristics of the multi-spectral camera focal plane, the imaging characteristics of the camera is analyzed. The large field view camera has multiple channels, and its MS and PAN images data volume is so big that the technology requirements of image registration, fusion, mosaic are relatively high.2) Image registration is a key pre-processing of image fusion and image mosaic.In the aspect of the image registration of the large field view camera single channel MS and PAN images, a registration method based on bilinear interpolation is proposed in the field of spatial correlation. Firstly, the multi-spectral images are magnified by image processing of bilinear interpolation algorithm. The MS image has the same size as the PAN image. Secondly, in the field of spatial correlation, a registration method based on bilinear interpolation is applied on the MS and PAN images. The experimental results show that, the proposed methods have higher rapidity, noise immunity and robustness, which makes the registration of the large field view camera obtain good results.3) In order to get the width of the image, the image of each channel has a seamless mosaic. This paper presents a multi-channel image mosaic method based on spatial cross-correlation, weighted average and moment matching. Firstly, the multi-channel images registration based on spatial cross-correlation makes the images quasi-aligned. Secondly, the weighted average algorithm fuses the registration image and eliminates splicing seams. Thirdly, the moment matching algorithm removes the band noise in the mosaic image, and the gradation of each channel achieves a uniform.4) A fusion method of MS and PAN images based on improved Pulse-Coupled Neural Network(PCNN) and region energy in Nonsubsampled Contourlet Transform(NSCT) domain is proposed. Firstly, the two original images are decomposed into a low frequency subband and more bandpass directional subbands by NSCT. Then, for the low frequency subband coefficients, an adaptive regional energy weighting image fusion algorithm is presented; while for the bandpass directional subband coefficients, based on improved PCNN, the algorithm uses the bandpass directional subband coefficients as the linking strength. After processing PCNN with the linking strength, new fire mapping images are obtained. The fire mapping image region energy is calculated, and the fusion coefficients are decided by the compare-selection operator with the fire mapping image region energy. Finally, the fusion images are reconstructed by NSCT inverse transform. In order to verify the effectiveness of the proposed fusion algorithm, respectively, using wavelet improved algorithm, Contourlet algorithm, NSCT algorithm and the proposed algorithm, for the same MS and PAN images, a fusion experiment is carried out. The evaluation and comparation of these algorithms experimental results are showed.5) In order to improve the color performance of the fusion image, a method of local color transfer based on Lαβ tansformation is proposed. The contrast with Reinhard global color transfer method is showed that the proposed method has good results.