基于生物视觉的颜色直方图特征提取  

Color histogram feature extraction based on biological vision

作　　者：张悦[1] 霍宏[1] 方涛[1] 

机构地区：[1]上海交通大学自动化系系统控制与信息处理教育部重点实验室,上海200240

出　　处：《高技术通讯》2014年第4期407-413,共7页Chinese High Technology Letters

基　　金：973计划(2012CB719903);国家自然科学基金委创新研究群体(X198144);国家自然科学基金青年科学基金(41101386)和国家自然科学基金(41071256)资助项目

摘　　要：根据人类视觉通路对颜色的感知过程,建立了颜色感知的层次性描述模型。用该模型模拟了自然光经眼睛进入视网膜(RETINA),到侧膝体(LGN),再到V1阶段的处理过程,最终在V1阶段的double-opponent神经元响应下形成具有方向信息的颜色特征。针对传统的颜色直方图包含信息较少,特征存在维数高、精度低的缺点,基于此模型,提出了描述具有对比度敏感、具有方向选择性和颜色恒常性的层次性颜色特征的方法。利用该特征对遥感图像进行了分类实验,实验表明了该特征比广泛采用的颜色直方图特征有更低的特征维数和更好的分类效果。对不同光照的图像进行了分类实验,验证了该特征具有颜色恒常性和更好的稳健性。According to the color perception of human visual pathways,a hierarchical model for description of color perception was established.By using the model,the neural processing for the natural light entering through the eye into the retina,the lateral geniculate nucleus （LGN）,then to the sytage of V1,was simulated,and eventually,a color feature with orientation information formed in the response of double-opponent neurons.As traditional color histogramfeatures have the shortcomings of less information,high dimension and low accuracy,a method for description of a hierarchical color feature with the contrast sensitiveness,direction selectivity and color constancy,was proposed based on this model.The experiment on remote sensing image classification was conducted by using the fenture,and the results show that the proposed method could achieve the better accuracy and the lower dimension than traditional ways.The classification of the images under different illuminations verifies that the proposed feature has the color constancy and better stability.

关 键 词：腹侧通路 层次性 double-opponent 颜色直方图 图像分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]