Haze Removal And Uneven Intensity Correction Of Images By Variational Methods

Sensors are affected by many factors when imaging, including the sensor itself, the illumination condition, the atmosphere and many other factors, which can cause degradation of the obtained images. Especially, remote sensing image is easily affected by the above factors. The observed remote sensing images are degraded, which can not meet the needs of practical application. In this paper we pay attention to the research of haze degradation and uneven intensity distribution. For the unevenness of atmosphere, images are affected by haze, which causes dynamic range compression and low contrast. Due to the illumination and the sensor itself, the observed scene appears uneven intensity distribution. Anyway, no matter the haze and the unevenness of intensity, they both cause the error between the obtained image and the actual object, which influences the following image analysis and interpretation works. Finally, it reduces data utilization, and even influences the overall remote sensing application. Thus, it is very important to remove haze and correct the uneven intensity distribution from the degraded images.The traditional correction algorithms are relatively complex and have limited extensions. Variational method has an advantage in image processing. Variational model has a unified framework, which can design the corresponding regularization according to different degradation problems. Therefore, to different degradation cases, this paper gives the suitable regularization prior, and then proposes the corresponding variational correction model according to the different physical degradation factor. The main contributions are in the following:A dark channel-based non-local variational method is proposed for haze removal and eliminating the interference of sensor blur and noise. The dark channel prior is that, some pixels have very low intensity and even zero in at least one color channel. But, considering sensor blur and noise in haze image, the dark channel prior is invalid. Thus, a three-stage algorithm for haze removal, considering sensor blur and noise, is proposed. In the first stage, non-local filter and non-local TV framework is used to preprocess the degraded image and eliminate the blur/noise interference to estimate the hazy image. In the second stage, we estimate the transmission and atmospheric light by the dark channel prior method. In the third stage, a combing the transmission regularization function is constructed. A regularized method is proposed to recover the underlying image. The steepest descent approach is used to solve the proposed variational model. Experimental results demonstrate the proposed method can remove haze while eliminating the blur/noise interference.A spatially adaptive retinex variational model for the uneven intensity correction of remote sensing images is proposed. According to the retinex theory, the relationship and the fidelity term between the illumination and reflectance are considered, which confirms that the revered result is not deviated from the observed image. The spatial information is considered, which is used to constrain the TV regularization strength of the reflectance. A spatially adaptive regularization weight parameter is constructed. In the edge pixels, a weak regularization strength is enforced to preserve detail, and in the homogeneous areas, a strong regularization strength is enforced to eliminate the uneven intensity. Also, the spatial smoothness of illumination and the "GW" criterion are considered. Finally, a spatially adaptive retinex variational model is proposed. Experimental results demonstrate the proposed method can correct uneven intensity distribution and preserve details. Compared to Li’s method and Michael’s method, the proposed method is better, based on the visual effect and quantitative assessments.A framelet-based sparse regularization model is proposed for the uneven intensity correction of remote sensing images. Considering the sparsity of reflectance, the analysis-based sparse regularization term is employed to constrain the reflectance. To illumination component, a L2norm is used to confirm the smoothness. Therefore, a framelet-based sparse regularization model is proposed. The alternating minimization algorithm and split Bregman method are adopted to solve the framelet-based sparse regularization model. The experiments, with both simulated images and real-life images, show that the proposed model can effectively correct the uneven intensity distribution and preserve the structure information.