【摘要】 针对特定类目标的识别,提出了一种优化的特征字典构造方法.根据特征原型的判别能力在不同属性参数上存在差别,采用了多轮训练的方法获得局部描述子分类能力在不同滤波器尺度上的统计分布,并据此对特征字典中具有不同滤波器尺度的特征原型的比例成分进行了调整,使分类能力强的特征原型在特征字典中占有更大的比例.实验结果证明,这样的字典构造优化方法和传统方法相比,可以得到更好的效率.更多还原

【Abstract】 Aiming at class-specific recognition tasks,a modified construction method for feature codebook building was proposed.As the codeword discriminability varies between different property parameters,we get the discrimination distribution upon different filter scales through a multi-run training phase,and modified the codebook content proportion for codeword at different filter scales,according to corresponding statistical distributions of discriminability.The test results demonstrate that the codebooks built with proposed modification achieve higher total-length efficiency.更多还原