Information Processing And Target Detection For Visible Blur Image

Generally, visible imaging utilizes the visible information of object reflection in order to realize detection imaging. The resolution of system imaging is very high. Achievement of acurate and rich object information, as well as precise observation, compose the two important parts of the application of visible imaging system, which play a significant role in the fields of theory and application. In the realm of practical application, visible imaging, who is based on face array imaging, may both tend to generate blurred images. Besides, complex natural scene can also exert very negative influence on the detection of object. From this, this dissertation do the following research works about the process of blurred information and the method of object detection for the blur and noisy image of visible imaging:In the study of image restoration, to overcome the shortcomings of traditional image restoration model and total variation image restoration model using single norm regularization, we propose a novel Hopfield neural network-based image restoration algorithm with adaptive mixed-norm regularization. The new error function of image restoration combines the L2-norm and L1-norm regularization types. A method of calculating the adaptive scale control parameter and the updating rule of neural network are also introduced. Experimental results demonstrate that the proposed algorithm is better than other algorithms with single norm regularization in the improvement of signal-to-noise and vision effect.In consideration of the motion blur and defocusing blur of images, a robust bispectrum-based method is proposed to estimate the blur parameters for the noisy images. According to the deduction of bispectrum of two blur types, bispectrum can not only reduce the influence of noise, but also correspond to a similar structure of blur function, which leads us to calculate the blur parameters easily by using the traditional computational formulas. In order to improve the identification accuracy, the curve fitting is used to obtain the functional relationship between the statistical characteristics in bispectrum and degraded parameter. The back propagation(BP) neural network trained by the above mentioned functional relations can identify the blur parameters. In the mixed blur of motion and defocusing blurs, the proposed method can also perform well.For the partial motion blur of visible TV, caused by motion object in static background, we establish the relationship among the object velocity, motion blur length, object range and camera parameter, such as exposure time. A novel algorithm of parameter identification for partial motion blur is proposed. The target range is known, while the motion direction of target is given by Fourier spectrum and Radon transformation. Then the motion direction is adjusted to horizontal. Through image segmentation, binarization, edge detection and self-correlation, the motion blur length is determined by the statistical information. Based on this, the object velocity is measured. The experimental results demonstrate the effectiveness of our method, even in the situation of complex object shape.The Wigner-Ville distribution(WVD) has a good characteristic that the images can be represented in spatial/spatial-frequency domains. For ground background images of visible imaging, through the pseudo Wigner-Ville distribution(PWVD) using one or two dimension window, the statistical property of Rényi entropy is studied. We also analyzed the change of Rényi entropy's statistical property in natural images when the man-made objects appear. From this foundation, a novel method for generating saliency map based on PWVD and Rényi entropy is proposed. After that, target detection is completed when the saliency map is segmented using a simple and convenient threshold method. For visible images, experimental results show the proposed method can effectively detect the military vehicles from complex earth background with low false alarm. The results of using one dimensional PWVD window are more accurate than those of using two dimensional window. Additionally, in the situation of adding noise and blur, this method can also perform a good property of robustness.