摘要
针对不完全小波树形结构分解提取纹理特征仅对清晰度高的图像效果佳,运算速度慢的问题,提出基于形态学预处理的不完全小波树形分解快速提取图像纹理特征的算法。首先采用形态学高帽—低帽变换对图像进行预处理,去除图像噪声,增强对比度;在提取纹理特征时,采用一致性判别;对于一致性强的图像,只利用图像的一部分进行不完全小波树形结构分解提取出能量、方向性等纹理特征,提高了运算速度;最后使用双概率神经网络(DPNN)的方法自适应地对纹理图像进行识别。利用Brodatz纹理库进行了仿真实验,并将该算法应用到了现场拍摄的海水中藻类细胞图像的识别。实验结果表明,该算法特征提取和识别速度快,尤其对于清晰度不高、现场拍摄的纹理图像具有较好的效果。
Low operation speed and being only fit for high quality images are the disadvantages of incomplete tree-structured wavelet.To deal with this problem,a new algorithm was proposed.Firstly,pre-processment using tophat-bothat was done to clear noise and to enhance contrast degrees;then the feature consistency was extracted.If its value was high,only one part of the image would be used in incomplete tree-structured wavelet.Otherwise,the whole image would be used.Lastly,Double Probabilistic Neural Network（DPNN） ws adopted here to identify images.Brodatz database was used for simulation,and the algae images pictured at scene was used as an application of this method.Result shows that this algorithm is fast in feature extraction and identification,with especially good performance at low quality images.
出处
《计算机应用》
CSCD
北大核心
2011年第6期1592-1594,共3页
journal of Computer Applications
基金
国家自然科学基金资助项目(60572064)
国家863计划项目(2006AA09Z178
2007AA09Z106)
关键词
高帽—低帽变换
不完全树形小波
一致性判别
双概率神经网络
tophat-bothat transform
incomplete tree-structed wavelet
consistency judgment
Double Probabilistic Neural Network（DPNN）
