基于网格量化和区间均值的动态目标识别

动态目标识别中，当摄像机和目标之间相对距离变化时，对于同一目标的多尺度序列图像。虽然目标区域图像大小和清晰度不同，但整体结构分布具有相似性和一致性。根据这一特点，提出基于目标区域网格量化的方法。在目标区域上形成采样网格，通过相对量化提取结构特征矩阵。对于同一目标序列图像中的每一个目标区域，其结构特征矩阵基本不变。结构特征矩阵反映象素分布的自相关性质。进一步，对由干扰引起的灰度值结构分布误差，采用网格区间均值的方法加以解决。目标区域网格量化是对相邻采样点的差值进行量化匹配。区间均值是对少数相邻采样点的进一步匹配。实例表明，本文的方法运算速度快、可靠性高，达到了实用目的。

Approach to Dynamic Object Recognition with Grid Unit Integers and Region Average

In dynamic object recognition, when the distance from the camera to the object changes, the overall constructions of the complex scale serial images of the same object are consistent and similar although the image magnitudes and articulation degrees vary within the object region. Starting from this feature, a method of dynamic object recognition is introduced. The method is based on grid quantization in the object region. Sample grids are formed in the object region. Feature matrixes are computed through relative quantization. The feature matrixes remain unchanging with respect to all regions of the serial images of the same object. The feature matrixes exhibit self-correlativity of picture element distribution. The region average method is used to erase distribution errors in the grey level structures caused by interferences. Grid quantization in the object region plays the part of quantizing matching among the D-values of adjacent sampling points whereas the region averages play the part of further matching among a small number of adjacent sampling points. Practices show that the proposed method is quick in computation and highly reliable in operation. The practicable purpose is attained.