In order to preserve the texture and edge information and to improve the space resolution of single frame, a superresolution algorithm based on Contourlet (NSCT) is proposed. The original low resolution image is transformed by NSCT, and the directional sub-band coefficients of the transform domain are obtained. According to the scale factor, the high frequency sub-band coefficients are amplified by the interpolation method based on the edge direction to the desired resolution. For high frequency sub-band coefficients with noise and weak targets, Bayesian shrinkage is used to calculate the threshold value. The coefficients below the threshold are determined by the correlation among the sub-bands of the same scale to determine whether it is noise and de-noising. The anisotropic diffusion filter is used to effectively enhance the weak target in the low contrast region of the target and background. Finally, the high-frequency sub-band is amplified by the bilinear interpolation method to the desired resolution, and then combined with the high-frequency subband coefficients after de-noising and small target enhancement, the NSCT inverse transform is used to obtain the desired resolution image. In order to verify the effectiveness of the proposed algorithm, the proposed algorithm and several common image reconstruction methods are used to test the synthetic image, motion blurred image and hyperspectral image, the experimental results show that compared with the traditional single resolution algorithm, the proposed algorithm can obtain smooth edges and good texture features, and the reconstructed image structure is well preserved and the noise is suppressed to some extent.
According to the theory of dark channel prior a image haze-removal algorithm is proposed in this paper. The algorithm uses maximum-minimum value filter combined together with guided filter to remove haze from the original image and uses wavelet to enhance the visual effect of the de-hazed image. Using maximum-minimum value filter only can cause the problem that the algorithm depending on the value of transmission lower limit excessively, by using maximum-minimum value filter combined together with guided filter the problem can be solved efficiently and the transmission matrix is refined adaptively. The white halos and patchy singularities which exist at the edge of the depth field in the reconstructed image is eliminated. Furthermore the algorithm refine the values of transmission which are estimated too big or too small. Finally wavelet is adopted to enhance the visual effect of the de-hazed image effectively. The objective evaluations of the reconstructed de-hazed image such as reconstructed image entropy, reconstructed image variance, reconstructed image mean square error, the degree of reconstructed image change and reconstructed image clarity are also studied in the paper, but these indicators can not represent the advantages and disadvantages of the performance of the image haze-removal algorithm, so it still needs further study in this field.