基于加权稀疏低秩组件编码的猪脸识别算法

Pig face recognition algorithm based on weighted sparse low-rank component coding

针对养殖行业中动物很难适应耳标的问题,采用非入侵的识别方式进行猪脸识别,提出了基于加权稀疏低秩组件编码的猪脸识别算法.应用视网膜皮层理论与区域协方差滤波器来估计光照,并结合文中新算法提出自适应伽马校正方法对获取的反射分量进行增强,以减少光照对识别结果的影响;同时,采用训练样本中的低秩组件构建字典矩阵,并重构残差函数处理误差,以提升算法应对含有污垢图像的识别性能.在JDD2017猪脸数据集上进行了光照和面部污垢验证试验,分别统计其识别率与耗时情况.结果表明:文中所提出算法显著优于传统稀疏表示方法,具有容忍光照变化、污垢和训练耗时短的优点.

To solve the problem that animals were difficult to adapt to ear tags in the breeding industry,the pig face recognition algorithm was proposed based on weighted sparse low-rank component coding by the pig face recognition in non-intrusive recognition method.The Retinex theory and the regional covariance filter were applied to estimate the illumination,and the proposed adaptive gamma correction method was used to enhance the reflection components to reduce the impact of illumination on recognition results.The low-rank components in the training samples were used to construct the dictionary matrix,and the residual function was reconstructed to process the errors and improve the recognition performance of the algorithm in dealing with the images containing dirt.The recognition rate and the time-consuming situation were calculated based on the light and facial dirt verification experiments on the JDD2017 pig face dataset.The results show that the proposed algorithm is significantly better than the traditional sparse representation method,and it has the advantages of tolerance to illumination changes and dirt with short training time.