Color Image Restoration Via Adaptive Local Kernel Regression With Structure Measurement

Affected by the illumination condition and set parameter, color image has been corrupted by weak blurring and strong noise. Meanwhile, the two distortions are ambivalent when we restored the color image. To solve this problem, the paper proposed an adaptive kernel regression method for image restoration. The proposed method transforms the image from RGB to YCbCr, and then we calculate the gradient matrix of Y component, with which we estimate the local covariance around each pixel. Then, we can establish the local Steering regression kernel for Y. Aiming to capture the local structure in image, we also introduce the Laplace convolution kernel, which can adjust the regression kernel adaptively by measuring the local structure. Based on the operation, the proposed method can make the two distortions mentioned above to achieve a reasonable balance. Experimental results show that, compared with the conventional methods, the proposed method not only suppressed the most noise but also has the stronger detail preserving and sharpening for real image.