基于机器学习的文字判别算法

Text Judging Algorithm Based on Machine Learning

复杂背景下的文字处理一直是OCR领域中的难题。基于SIFT、K-means和SVM三种算法相结合,提出了一种全新的复杂背景下的文字判断方案。在学习阶段利用SIFT进行特征点提取和描述,并用K-means进行聚类,最后用SVM进行学习。在测试阶段对图片进行了SIFT特征点提取,并用学习阶段得到的聚类中心和SVM判决函数进行最终的判断,得出图片中是否含有文字的结果。

Text processing under complicated background is always the difficult problem in OCR domain. This paper based on SIFT, K-means and SVM proposes a brand new method for text judging under complicated background. In the training phase, the sift detector and sift descriptor is used to get the eigenvectors of the pictures. Then these eigenvectors are clustered by using K-means and trained by using SVM. In the test phase, the key points and eigenvectors of a picture are acquired by using SIFT. And then using the cluster center and SVM judge function, the final judgment is implemented, thus the conclusion whether there exist texts in the picture is reached.