The Study Of Remote Sensing Image Fusion Method

The biggest characteristic of remote sensing is that image information is obtained from multi-sources. Due to the multi-level platform, different remote sensing platforms have different height, speed, observation area and image resolution. Different remote sensing platforms can provide different spectral resolution, different time resolution and different spatial resolution remote sensing images in the same area, This paper studies the fusion method between high spatial resolution remote sensing image-SPOT and multi-spectral remote sensing image-TM.There is pretreatment, including filtering and image enhancement on image before fusion. Image matching methods based on reference points and template matching are used. The main purpose of remote sensing image fusion is that two images are fused according to certain rules and image with high spatial resolution and extensive spectral information finally acquired. There are many image fusion methods. Early fusion methods are pixel-level fusion, such as HSI transform, weighted average, PCA, high-pass filtering and wavelet transform. HSI transformation, PCA principal component transform are all ingredients alternative method. Its fusion characteristics are: spatial detail information is enhanced for fusion multi-spectral image, , but leads to a large spectrum distortion.This paper focuses on multi-resolution analysis based on wavelet transform in image fusion. This article studies a large number of comparative analysis using different wavelets and different decomposition level . Two methods are introduced in this article: direct substitution method and entropy-based fusion method. Through integrating wavelet decomposition coefficients and wavelet inverse transform, image with spectral information and high-resolution can obtained. Then use other fusion rules to do comparisonBecause the traditional decomposition methods based on pyramid decomposition and wavelet decomposition methods are not well suited for high frequency band contains a large amount of information, we put forward a discrete wavelet packet transform on region image fusion . Binary, grayscale ratio calculation are used to determine which method to use in specific region. Finally inverse wavelet packet transform will be used to reconstruct the fused image. Experiments show that this method can effectively improve the quality of image fusion.Evaluation of the fusion effects, in addition to directly subjective visual interpretation, the article uses the following method to evaluate the effect of image fusion: information, statistical characteristics, correlation, gradient value . But for image with obvious features, evaluation is most on subjective visual interpretation.