一种用于图像编码的区域分割新方法

A Novel Region Segmentation Algorithm with Neural Network for Segmented Image Coding

为了适应图像分割编码的需要,提高编码性能和效率,本文研究了一种图像区域分割新方法.源于人眼成像原理和视神经网络的知觉分割特性,首先提出一种具有脉冲耦合和梯度锐化能力的神经元网络模型.然后通过构造一个拟合函数对相邻神经元的相似刺激输入进行平滑处理,而对具有不连续变化特性的刺激输入进行锐化,使得神经元比较容易地感知到均匀亮度区域和目标边缘的准确位置.最后通过实验验证了该算法的有效性.本文算法能够准确、有效地的分割出均匀区域,并且与原始图像具有很好的对应关系.在将本文算法应用到图像区域分割编码中,能够大大提高编码的效率,并得到高质量的重建图像.

A novel image region segmentation method is proposed to improve the efficiency and characteristic for segmented image coding. Based on human visual system and perceptive segmentation, a neural network with pulse-coupled and gradient-sharp- ened is introduced firstly. Then, a fitting function is used to smooth the neighboring similar stimulus and sharpen the non-continuous. In this way,neurons could find the accurate location of uniform regions and objective boundaries.At last,experimental results show the efficiency of the algorithm. The homogenous regions can be segmented from its background effectively and accurately. In addi- tion, the segmented regions are corresponding to objects in scene. There are high-quality constructed image if this segmentation algo- rithm is adopted by segmented image coding.