针对图像中不包含明显直线的情况,提出一种基于特征点提取的图像倾斜校正算法。该算法建立在与无倾斜的训练图像比对基础上,利用特征点构造直线,不依赖于原图中是否存在直线,具有尺度、平移无关性。首先使用双向最大相关系数匹配,匹配正确率较高;然后利用大数原理对数据进行处理,去除误匹配的影响,该算法最少可以利用两个匹配对,即可检测出图像倾斜角度。结合应用背景,还设计了一种用于特征提取的圆形模板,具有类似于旋转不变的性质。

For images without obvious lines, a novel skew correction method based on feature points matching is proposed. In this method, the correction does not use actual lines, but relies on the lines constructed by the feature points extracted from the images. It is translation and scale invariant. Through bidirectional best correlative coefficients calculation, a high matching rate can be achieved. Mismatching can be decreased, by using the method based on the Law of Large Numbers. It can get the correct slant angles by using only two matching pairs. In addition, due to the special application background of this paper, a disk template is designed to extract the Eigenvector of the features, which leads to rotation invariance.