基于极坐标的树木叶片特征提取方法

摘要

提出了一种在极坐标下进行特征提取的方法，并将其应用于树木叶片图像识别中。该方法首先将目标图像二值化并映射到极坐标下，提取具有旋转、缩放、平移不变性的曲线面积比率、跨度比、饱和度和高度变化率等特征。然后通过最小欧式距离对叶片进行计算、识别。该方法克服了直角坐标系下特征提取方法计算量大，花费的时间长的缺陷。基于该方法的叶片特征提取算法的时间复杂度为O(n)。对于叶型有区分，且需要快速进行树叶识别的树叶数据集有很好的效果。该方法在常见的13种树木叶片中进行测试，平均正确识别率达到90%以上。%The leaves feature extraction method based on polar coordinates is proposed in this paper. The proposed approach is mapping the binary image to polar coordinate and converting them into a curve firstly, then extracting some invariant from the curve. Finally, minimum Euclidian distance is used to train and recognize the leaves. The experiment proves that the approach has the smaller calculation and a good recognition effect. This method is suitable for the leaf shape which has distinction, and is used to identify the leaves needed in the short time. 13 species of leaves have been tested in this method, and the average correct recognition rate is greater than 90%.