Algorithm Analysis And System Implementation Of Text Judging Recognition Based On Machine Learning

Text processing under complicated background has always been a difficult problem in the OCR domain. This paper proposed a brand new algorithm of text judging under complicated background based on the previous algorithms and proved that our algorithm has high accuracy and speed.In this paper, we first introduce the concepts of image feature, clustering algorithm, and data mining, and then we introduce the SIFT, k-means, SVM algorithms. We also introduced the characteristic of these three algorithms. Then, we proposed a brand new algorithm of text judging under complicated background based on these three algorithms. In the training phase, this paper uses sift detector and sift descriptor to get the features of the pictures. Then these features are clustered using k-means and trained using SVM. In the test phase, we first get the keypoints and features of a picture using SIFT. By using the cluster center and SVM judge function, we get the conclusion of whether the there are texts in the picture.Furthermore, we have done lots of experiments based on our algorithm, including the scale, rotation of the images, the real images and proposed special text judging and recognition algorithm under complicated background based on machine learning. Also, we replace the traditional k-means with a new k-means based on our algorithm in our experiments and get a better result. With lots of experiments, we proved the feasibility and high accuracy of our algorithm.In the end, we have designed a J2EE system based on our algorithm and added the OCR module and feedback system. Also we have tested the performance of our system and proved that this system is fast with high performances.