论文部分内容阅读
常规图象边沿提取算法多基于微分滤波的思想,由于噪声影响,难以取得定位与抗噪的很好折中。本文应用模式分类的概念,选择抗噪能力强的空间矩作统计特征,设计了一个简单的非线形神经网络边沿算子。利用网络学习的能力来直接估计后验分类概率密度,一步完成从灰度图到边沿图的映射,对输出采用固定门限而无需后处理,即可获取单一宽度的边沿。文中给出了合成图与真实图的对比实验结果,表明本算子具有良好的定位与噪声抑制的能力,具有较高的实用价值。
Conventional image edge extraction algorithm based on the idea of differential filtering, due to the influence of noise, it is difficult to obtain a good compromise between positioning and anti-noise. In this paper, by applying the concept of pattern classification and selecting the space moment with strong anti-noise ability as the statistical feature, a simple non-linear neural network edge operator is designed. The ability of network learning is used to directly estimate the posterior classification probability density. The mapping from grayscale to edge map is completed in one step, and the output is fixed threshold without post-processing to obtain single-width edge. The experimental results of the composite graph and the real graph are given in the paper, which shows that this algorithm has a good ability to locate and suppress noise. It has high practical value.