Research On Methods Of Dehazing Optical Remote Sensing Images

Optical remote sensing images have been widely used in many areas including both national defense and economic construction, but a very large percentage of images acquired are contaminated and look illegible due to that the optical imaging sensor is easily effected by haze in the atmosphere. As a result many hazy images are not applicable in practice. So that research on methods for haze removal is necessary to improve visualization of the hazy image in fact. Methods involving homomorphic filtering, multi-scale and image restoration are mainly studied to process hazy images in this thesis. The main works completed are as follows:1. Models for the general image degradation and hazy image are summarized. Properties of the hazy image are analyzed thoroughly from three aspects consisting of spectrum, spatial distribution and frequency.2. In order to resolve the problem of processing the color hazy image using homomorphic filtering with the color space transformation, the way of introducing homomorphic filtering with HSV transformation is put forward since the transformation of HSV is better than that of HSI in some aspects.3. According to the means of removing haze from the image based on discrete wavelet transform, sub-images in high frequency are deal with a specific plan, but the plan for the sub-image in low frequency is uncertain. So a homomorphic filter is designed for processing the sub-image in low frequency. Methods for processing the hazy image based on stationary wavelet transform are investigated extensively by employing six plans to deal with sub-images in low and high frequency. The result shows that the best way to process the hazy image is to filter the non-hazy area of the sub-image in low frequency and enhance the sub-images in high frequency with a pre-defined filter and non-linear function respectively.4. A method for the single image dehazing using dark channel is carried out. An issue is discussed when dark channel is invalidated. It's proposed to estimate the atmospheric light for images got near ground by dividing an image into some blocks, and color images of the satellite in the blue channel. It's suggested to have a pre-process step of white balance prior to the haze removal algorithm because the fog may be not pure white in the image at some time. Gamma correction is performed for the resulting image obtained by the haze removal algorithm which looks too dark.5. A kind of method for haze removal from the image based on median filter is analyzed and implemented.6. All methods described above for haze removal are integrated into one engineering system, doing well in removing haze from both color and grey images and improving visualization of the hazy images greatly.