Research On Key Algorithms Of Image Sharpening Based On Video Surveillance

With the increasing popularity of the image forming apparatus and the mounting demand for security of people, the using of video surveillance in daily life becomes more and more widespread, and the resulting phenomenon of image degradation become increasing. In the process of imaging, images may produce different levels of degradation by reason of various factors. However, many applications need to get high quality images. How to recover the degraded image effectively or to enhance the image effectively detail we needed, become an important research topic in image processing. The relative motion between the camera and the object produced the motion blurred image. Caused by the numbers of suspended particles in the foggy day, the image is low contrast and fuzzy. This thesis mainly research the recovery algorithm of motion blurred images and foggy images, and the enhancement algorithm for image texture, because the two phenomenon are often encountered in daily life.The main work of this thesis is lists as follows:(1) In this thesis we fully research the relationship between the spectrum of motion blurred image and the parameters of the point spread function (PSF). The accuracy of the estimation of parameters of the motion can impact on the result of image restoration, so this thesis proposed the modified Radon transform on the basis of block binarization for the estimation of the motion’s parameters. The motion direction can be estimated accurately by the use of this method, regardless of whether the image has noise. On the basis of the accurate estimation of the motion direction, the distance of motion can be estimated more effectively by the use of the method of self-related differential operations. The accuracy of the PSF can be improved by use of the two parameters mentioned above. The experimental results show that the algorithm can better improve the quality of the degraded image.(2) This thesis research the adaptive enhancement algorithm based on the fractional differential. Based on the Gabor wavelet-based texture matching and the approach to select optimal parameters based on the Logarithmic Michelson contrast measure by entropy and PLIP (LogAMEE), the adaptive fractional differential algorithm on the strength of Gabor features was proposed in this thesis. This method can determine adaptively the optimal order of fractional differential for some texture image. However, the images with same Gabor features sometimes have texture of different complexity. So another parameter called fractal dimension which can characterize the texture complexity was introduced to make the proposed algorithm more stable, so an adaptive fractional differential algorithm for image enhancement based on the Gabor features and the fractal dimension was proposed in this thesis. This method can adaptively determine the order of fractional differential according to its own image texture features. The experimental results show that both methods can adaptively enhance the image based on image characteristics by use of fractional differential, and the latter has better stability.(3) This thesis research the algorithm of image defogging based on the dark channel prior, and the effect of this algorithm is well. But when there are the presence of sky or object in white in the image, its effect will become worse because this image does not meet the principle of the dark channel prior. Therefore, a novel defog algorithm is studied and proposed, which based on the logarithmic Michelson contrast measure by entropy (AMEE). The core of the algorithm is still dark channel prior information, but the estimation of transmittance introduce the AMEE contrast which can enhance the contrast of image while reducing the loss of the image information. This approach improves the defogging effect of a class that contains an area of sky, at the same times it takes less time on the estimation of transmittance. So this algorithm can improve the efficiency of defogging.