The Key Algorithms Research On Text Detection In Natural Scene Images

In the stage of information explosion, people get information more and more relying on the content-based retrieval technology. Image can contain rich information, and the text information in that is usually particularly important. Text extraction in images has become a hot research topic in computer vision research field recently. As a key component of text extraction technology, text detection is very important. However, Due to the complexity of background, the uncertainty of text size and vulnerable to light and occlusion, text detection in natural scene images is a very challenging problem.In this paper, it improves the algorithm of text detection in natural scene images on the following aspects: candidate text connected-component(CC) extraction, designing of Chinese character's feature and character grouping. a)In the method of extracting maximally stable extremal region(MSER) as candidate text CC, it prunes the nested MSER for the precision of classifying text and non-text CCs and reducing calculation by using statistical feature. b)In the Chinese character classification stage: after careful observation of the strokes of Chinese character's structure, it proposes a new CC regulation feature based on the CC's skeleton to classify text and non-text CCs. c)In the character grouping stage: this paper proposes a grouping method based on C4.5 decision tree on the foundation of traditional heuristic rules. It can get the value of the corresponding parameters and more effective grouping rules by means of learning from samples.By the improvement of text detection algorithm in above aspects, simulation experiment on the commonly used data sets shows that: the method proposed performances well in the images which have complex background and uneven illumination's interference, and improve the precision and recall rate of text detection in nature scene images significantly.