2D非调和分析耦合扩散张量的图像修复算法

Image Inpainting Algorithm Based on 2D Non-harmonic Analysis Mechanism and Diffusion Tensor

中文关键词: 图像修复  2D非调和分析  扩散张量  成本函数  合成纹理  块效应

英文关键词: image inpainting  2D non-harmonic analysis  diffusion tensor  cost function  synthesis texture  blocking effect

中文摘要:

      目的 针对当前图像修复算法在面对复杂纹理的高频成分时， 修复区域残留一定的块效应与纹理不连续性， 产生较大失真等难题， 提出了2D非调和分析机制耦合扩散张量的图像修复算法。方法引入傅里叶变换机制， 联合最速下降法与牛顿法， 设计2D非调和分析机制； 提取损坏区域初始轮廓，嵌入扩散张量， 增强了损坏区域周围的像素信息表达能力， 构建新的优先级计算模型， 确定出优先级最大的先验块及其填充顺序； 基于2D非调和分析， 嵌入权重因子， 定义成本函数， 提取该先验块的精确频谱， 通过重构2D波形， 合成新纹理， 并将该纹理传播至损坏区域， 完成修复。结果 该算法与当前性能较好的图像修复技术相比， 在大面积高频分量损坏图像复原中， 具有更高的修复质量与结构相似度， 很好地避免了目标区域的不连续纹理与块效应。结论 该算法能够较好地修复大面积高频损坏区域的图像。

英文摘要:

      Targeting at the drawbacks of the current image inpainting algorithm in processing high-frequency component images with complex textures, such as the obvious blurring effect and blocking effect in the repairing areas, leading to serious distortion, an image inpainting algorithm based on 2D non-harmonic analysis mechanism and diffusion tensor was proposed in this paper. By introducing the FFT mechanism, the 2D non-harmonic analysis mechanism was designed by combining the steepest descent method and Newton′ s method. The priori block with maximum priority and the filling order were determined by constructing the new computing model of confidence term and data term based on the enhanced pixel information presentation capacity in the surrounding of the damaged areas through extracting the initial contour of damaged areas and embedding diffusion tensor. Then the cost function was defined by embedding weighting factors and the precise frequency spectrum of the priori block was extracted based on 2D non-harmonic analysis. The new texture was synthesized by reconstruction of 2D waveform, and propagated to missing area to accomplish the inpainting. Simulation results showed that: in comparison with current inpainting algorithm having good performance, this algorithm had higher inpainting quality and structural similarity for effectively eliminating the discontinuous textures and blocking effect in repairing area in inpainting of images with large-area high-frequency component damages. In conclusion, this algorithm could repair the images with large-area high-frequency damages.