Concerning the poor quality of existing image super-resolution reconstruction caused by single dictionary, a new single image super-resolution algorithm based on classified image patches and image cartoon-texture decomposition was proposed. Firstly, an image was divided into image patches which were classified into smooth patches, edge patches and texture patches, and the texture class was divided into cartoon part and texture part by Morphological Component Analysis (MCA) algorithm. Secondly, ege patches, cartoon part and texture part of texture patches were applied respectively to train the dictionaries of low-resolution and high-resolution. Finally, the sparse coefficients were calculated, then the image patches were reconstructed by using the corresponding high-resolution dictionary and sparse coefficients. In the comparison experiments with Sparse Coding Super-Resolution (SCSR) algorithm and Single Image Super-Resolution (SISR) algorithm, the Peak Signal-to-Noise Ratio (PSNR) of the proposed algorithm was increased by 0.26 dB and 0.14 dB respectively. The experimental results show that the proposed algorithm can obtain more details in texture with better reconstruction effect.

The difficulty in sketch-based image retrieval is the effective recognition of images with different scales, positions, rotations and deformations. In order to identify and retrieve images of different scales, positions and rotations more accurately, a Sketch-Based Image Retrieval method Using Local Geometry Moment Invariant (SBIRULGMI) was proposed. Firstly, the geometric characteristics of image were used to determine the coordinate system of image. Secondly, the geometry moment invariant of image blocks which were divided averagely based on the generated coordinate system was calculated to form a eigenvector. Then, the similarities between query sketch and images in database were calculated based on Euclidean distance. Finally, the retrieval results were obtained from the similarity ranking and optimized according to Ant Colony Optimization (ACO). Compared with Shape Context (SC), Edge Orientation Histogram (EOH), GAbor Local lIne-based Feature (GALIF) and MindFinder, the retrieval accuracy of the proposed method in image database of MPEG-7 shape1 part B was increased by 17 percentage points on average. The experimental results show that the proposed method not only has a better recognition effect on the images after translation, scaling and flipping transformation, but also has better robustness to a certain degree of rotation and deformation.

To resolve the problem that more distortions occured in image retargeting algorithm, an image retargeting algorithm based on the Parallel Translation of Gridlines (PTG) was put forward. Firstly, Achanta algorithm was used to compute the important degree and extract the main object. Secondly, the optimal grid line displacement was calculated. Using grid lines movement can keep the size of important areas and protect the aspect ratio of object and the dual constraints can avoid distortion. At the same time, the lower and upper thresholds were used to restrain the distortion caused by excessively narrowing and widening grids. Finally, in order to achieve better effect, a edge discarding process was introduced to assign wider space to the important area for reducing the distortion. Image retargeting survey system was used to compare PTG with the methods including column removal method with importance diffusion, seam carving method with importance diffusion and grid warping method, and PTG got a better result in images with obvious main goal. The experimental results show that PTG not only has less distortion but also retains the interest area and important object of the image than the comparison methods.

For resolving the problem of curvature calculation for scattered point cloud data with strong noise, a robust statistics approach to curvature estimation was presented. Firstly the local shape at a sample point in 3D space was fitted by a quadratic surface. In addition,the fitting was performed at multiple times with randomly sampled subsets of points, and the best fitting result evaluated by variable-bandwidth maximum kernel density estimator was obtained. At last, the sample point was projected onto the best fitted surface and the curvatures of the projected point was estimated. The experimental results demonstrate that the proposed method is robust to noise and outliers. Especially with increasing noise variance, the proposed method is significantly better than the traditional parabolic fitting method.