The enhancement of remote sensing images by local brightness balancing

Because the traditional image enhancement methods such as linear contrast stretching or nonlinear histogram equalization use the global histogram as a reference to starch the image gray levels, some local or detailed variations of ground objects on the images usually will be over enhanced or suppressed. In fact, some of studies based on local brightness adjustment were proposed to improve the drawbacks mentioned above. For example, the Multi-Scale Retinex (MSR) algorithm assumes that the image is modulated by a spatially changed illumination and tries to eliminate this effect at local scale. However, in the case of multi-spectral image, when MSR is applied to each individual band, color distortion can be caused due to the brightness variations are different from band to band. In order to enhance the original multi-spectral image and keep its color characteristic at same time, the MSR was only apply to the intensity component extracted from original multi-spectral image. In this case the intensity component was usually obtained by RGB to ISH transformation or Principle Component Analysis (PCA). However, the intensity component of RGB to ISH transformation generally has problem to capture the most important brightness variation in multi-spectral image. On the other hand, PCA can have better performance in the extraction of brightness variation in its first principle component. In this study, a PCA based MSR algorithm is applied to enhance a high resolution (8 meters) and large area (200 km x 400 km) image mosaic obtained by FORMOSAT-2 satellite for testing its performance. Experiment results show that the PCA based MSR can provide better enhanced results when it is compared to traditional percentage clip linear stretch methods.