Abstract: The feature point detection plays an important role in the sequential process of image analysis and understanding. This paper proposes a new method of image feature point detection, which is based on the mechanism of visual receptive field and information flow feedback. By using the simple photoreceptor cell layer of the receptive that has a self-adaptive structure, a gross detection of the feature points of Gaussian differences acquired with convolution operation is conducted; redundant points are removed with the feedback mechanism of pulse information flow, and representative feature points are finally obtained under the visual attention mechanism. Although there are significantly fewer final feature points in the new algorithm than the traditional, when the image is rotated by 30°, 60°, 90° and the scale transformation is 0.8, 0.9, 1.1 and 1.2 respectively, image feature points in the new algorithm show more stable consistency. The method of feature point detection discussed in the paper provides a brand-new and effective idea for image processing based on visual physiological characteristics.

Key words: feature points detection, self-adaption receptive field, feedback of neurons, visual attention mechanism

摘要： 特征点检测性能对于后续图像分析和理解起着关键的作用，基于视觉感受野以及信息流反馈等视觉机制，提出了一种图像特征点检测新方法。利用感受野自调节特性构造简单细胞感光层，对卷积运算所获取的高斯差异结果进行特征点粗检测；利用脉冲信息流的反馈机制进行冗余点的剔除，最终获得视觉注意机制下的代表性特征点。在图像旋转角度为30°、60°、90°，尺度变换因子为0.8、0.9、1.1和1.2时，新方法在最终特征点数量均显著少于传统算法的情况下，图像特征点一致性稳定性结果较优，该方法将为生物视觉机制及其在图像处理中的应用提供崭新而有效的思路。

关键词: 特征点检测, 自调节感受野, 神经元反馈, 视觉注意机制