基于视觉特征信息量度量的高斯尺度参数自适应算法

An Adaptive Algorithm of the Gaussian Scale Parameters Based on the Scale-Space of Visual Feature Information

为了避免计算过于复杂或因丢弃过多关键信息而造成失真过大的问题,在高斯尺度空间的构造中应正确选用尺度参数,以使图像信息的变化呈现均匀的特点。目前,许多高斯尺度空间应用中采用的层之间的尺度参数关系并不明确,使得分层效果不理想。本文基于视觉特征模型,提出一种自适应高斯尺度参数的算法,并通过对SAR图像降噪处理对比试验验证了它的有效性,从而为图像的高层次处理如目标识别等提供了信息量稳定变化的尺度空间。

For the purpose of avoiding the problem of complicated computation or over-distortion because of losing too much key information, it is crucial to choose appropriate scale parameters during the construction of the Gaussian scale- space in order to represent the image information in uniform distribution. At present, the scale- parameter relations between the layers of the Gaussian scale-space in many applications is not clear, which may lead to bad effect on the layers. The paper proposes an adaptive algorithm of the Gaussian scale parameters based on the scale-space of visual feature information. The method is evaluated by a SAR image denosing experiment in the last section. The experimental results provide the uniformly distributed information in the scale-space which is useful in higher-level image processing such as object recognition.