Uneven Intensity Correction Of Remote Sensing Images By Variational Methods

Remote sensing sensors aboard are disturbed when imaging by the in-situ conditions of themselves, the illumination, the atmosphere and the landscape, which degrades the quality of the observed images. One of the common degradations is the unevenness of the intensity distribution. According to the influencing scope, the unevenness can be divided into two types, the overall unevenness and the local unevenness. The overall unevenness is attributed to the synthesis many in-situ conditions, in which the sensor and the illumination play the major roles. In a scene with overall intensity unevenness, objects belonging to the same class exhibit saliently different intensities. The local unevenness is mainly caused two factors, the cloud (including the thick cloud, the thin cloud and the haze) and the shadow. The former leads the intensity increase due to the complicated atmospheric scattering. The latter leads the intensity decrease due to the obstruction the incident light. No matter the overall and the local unevenness, they both cause the error between the observed intensity and the actual intensity, which influences the human interpretation and the following image analysis and process. Therefore, it is an essential work to correct the unevenness of the remote sensing images for the theoretical research and the practical applications.The physical mechanism for each kind of unevenness is different, so that the correction model should be designed according to the corresponding imaging characteristics. Traditional correction methods have distinct forms and limited extensions. To different types of unevenness, this dissertation intends to construct variational correction models according to the corresponding physical characteristics. The main research contents are as following.A perceptually inspired variational method (PIVM) is proposed for correcting the overall uneven intensity of remote sensing images. The PIVM is characterized by human visual system properties, locality, color constancy and non-linearity. The proposed method shares the same intrinsic scheme as the Retinex theory, but the reflectance in this method is solved directly within the limited dynamic range and is supposed to comply with the gray world assumption. Considering the smoothness of the illumination, the H1half norm prior is taken to constrain it; and the complexity of the reflectance, the proposed method integrates H’half norm and total variation priors to inflict varying constraints on homogeneous and heterogeneous regions adaptively. The minimum of this variational model is calculated using the steepest descent approach. Experimental results suggest that the PIVM can correct the overall intensity unevenness, constrain the dynamic range of the result and enhance the contrast. Comparing with the original variational Retinex model, the PIVM yields more even result and overcomes the undercorrection on regions with high intensity and overcorrection on regions with low intensity.A new variational gradient-based fusion method for visible and short-wave infrared (SWIR) imagery is proposed to enhance edges and textures, and meanwhile remove the thin cloud and haze in visible imagery. The atmospheric scattering among channels with different wavelengths is distinct from each other. The SWIR wavelength is long enough to round the particles of the thin cloud and haze, so that the corresponding imagery is clear and sharp. Meanwhile the SWIR imagery is linearly correlated with the visible imagery, which has been demonstrated. The proposed method employs the above mentioned characteristics of SWIR imagery to integrate gradient information from it with the visible imagery. The fusion result is a single image with true color and sharp gradients. A constraint based on band correlation is included to improve the enhancement and implement dehazing. The band correlation is according to the quantitative relationship between the wavelength and the atmospheric effect caused by Rayleigh scattering. The mean haze projection (MHP) is constructed to constrain the iteration process of the variational gradient fusion. In this study, both clear and hazy Landsat ETM+images are used in the experiments. By visual assessment, the gradient of the fused image is more salient than that of the original image, and the true color is well preserved. With the inclusion of the MHP, the proposed fusion method yields almost haze-free results and the true spectral information is well restored. Quantitatively, the Metric Q of the fused images is significantly raised comparing with that of the original images. A new shadow removal method by using nonlocal (NL) operators is proposed for remote sensing images with high spatial resolutions. Shadows are evident in most images with high resolutions, particularly in urban scenes, and their existence obstructs the image interpretation and the following application, such as classification and target detection. Most current shadow removal methods were proposed for natural images, whereas shadows in remote sensing images show distinct characteristics. We have therefore analyzed the characteristics of shadows in aerial images, and in this part, we propose a new shadow removal method for aerial images. In the proposed method, the soft shadow is introduced to replace the traditional binary hard shadow. NL operators are used to regularize the shadow scale and the updated shadow-free image. Furthermore, a spatially adaptive NL regularization is introduced to handle compound shadows. The combination of the soft shadow and NL operators yields satisfying shadow-free results, preserving textures and holding regular color. Different types of shadowed aerial images are employed to verify the proposed method, and the results are compared with two other methods. The experimental results confirm the validity of the nonlocal regularized shadow removal method and the advantage of the soft-shadow approach.