The key to the restoration of rotational motion blurred image is how to restore the image under a low cost and to correct the irreversibility of the degradation function matrix. Based on the special qualities of degradation function matrix and precise deduction in space-domain, we present a new approach using gradient-loading for restoration of rotational blurred image. By easily adding a gradient operator, the irreversibility of the original matrix is corrected and can be applied for inverse filtering then. Gradient-loading is the optimized approach which combines the advantages of both the approaches using constrained least square filtering and traditional diagonal-loading. Compared with the approach using least square filtering, its peak signal-to-noise ratio (PSNR) is improved from 3.18 to 6.46 dB, while the computing time is reduced to 1/2-1/3. Experimental results demonstrate the effectiveness, noise-resistibility, robustness, and low complexity of this approach, which make it more suitable for real-time environment.

In the real-world application of face recognition system, owing to the difficulties of collecting samples or storage space of systems, only one sample image per person is stored in the system, which is so-called one sample per person problem. Moreover, pose and illumination have impact on recognition performance. We propose a novel pose and illumination robust algorithm for face recognition with a single training image per person to solve the above limitations. Experimental results show that the proposed algorithm is an efficient and practical approach for face recognition.

Conventional Gabor representation and its extracted features often yield a fairly poor performance in extracting the invariance features of objects. To address this issue, a global Gabor representation method for raised characters pressed on label is proposed in this paper, where the representation only requires few summations on the conventional Gabor filter responses. Features are then extracted from these new representations to construct the invariant features. Experimental results clearly show that the obtained global Gabor features provide good performance in rotation, translation, and scale invariance. Also, they are insensitive to illumination conditions and noise changes. It is proved that Gabor filters can be reliably used in low-level feature extraction in image processing and the global Gabor features can be used to construct robust invariant recognition system.

针对常规综合鉴别函数制作的复合匹配滤波器，对不同旋转角度的训练样本输出的相关峰值波动性较大，不便于后续探测和处理等问题，对常规综合鉴别函数到编码匹配滤波器的编码转化过程进行分析，找到导致相关峰值波动的原因，提出改进方法，即在计算综合鉴别函数的线性组合系数时将编码运算也考虑进去，并根据滤波器对各训练样本相关峰值的倒数来确定线性组合系数。计算机仿真结果表明：改进后的方法降低了复合滤波器对不同旋转角度样本输出相关峰值的波动性，提高了相关峰自动判别的准确性。

Through deducing the relationship between support vector machine (SVM) and correlation principle, the optimal hyperplane is proved as a correlation filter when the kernel function is the linear kernel. So a new correlation filter, named linear SVM correlation filter (LSCF), is proposed. The filter has not only shift-invariance, but also SVM properties. The real images of laser radar are used as experiment data, and LSCF is used to solve the in-plane rotation invariance. The results show that the filter can recognize the different rotated objects, and the correlation output is stable. The filter is insensitive to the noise and gray change, and has good discrimination ability. In the same design way, LSCF is also suitable to solve other problems of correlation distortion.

Texture analysis is a basic issue in image processing and computer vision, and how to attain the rotation-invariant texture characterization is a key problem. This paper proposes a rotation-invariant texture analysis technique using Radon and Fourier transforms. This method uses Radon transform to convert rotation to translation, then utilizes Fourier transform and takes the moduli of the Fourier transform of these functions to make the translation invariant. A k-nearest-neighbor rule is employed to classify texture images. The proposed method is robust to additive white noise as a result of summing pixel values to generate projections in the Radon transform step. Experiment results show the feasibility of the proposed method and its robustness to additive white noise.

In order to resist geometric attacks, a robust image watermarking algorithm is proposed using scale-invariant feature transform (SIFT) and Zernike moments. As SIFT features are invariant to rotation and scaling, we employ SIFT to extract feature points. Then circular patches are generated using the most robust points. An invariant watermark is generated from each circular patch based on Zernike moments. The watermark is embedded into multiple patches for resisting locally cropping attacks. Experimental results show that the proposed scheme is robust to both geometric attacks and signal processing attacks.

The target recognition of laser radar becomes hot research in recent years, because laser radar can produce high space resolution and collect rich target information, such as range image, intensity image and Doppler image. In the vertical detection of laser radar, the problem of in-plane target rotation invariance is firstly solved. In the paper, a new support vector machine (SVM) correlation filter is presented, which simultaneously has the attractive attributes of SVM and common correlation filter. Exploiting the idea of margin of separation maximization, the design criterion is produced. The filter is synthetic by the multiple training images which are generated by rotating one image. The real range images of laser radar are used to finish the correlation experiments. The results show that the filter is not sensitive to the noise, the correlation peak is changed slightly for the different testing images, and the precision of location is high. This design way can be used in other recognition fields.