A Novel Text Location Algorithm In Complex Color Images

In this paper, a Connected Component-based Stroke Width Transform algorithm (CCSWT) is proposed, and it is combined with sparse representation and conditional random field (CRF) algorithm to locate texts in color images. Firstly, a color quantization algorithm is carried out to convert a color image into binary image levels. Second we select character candidates in the each binary image respectively and merge them into text line candidates. In every image level, sparse representation is used to obtain the probability of a connected component (CC) being character, and at the same time CCSWT is used to calculate the stroke width of the CC. Then we input the stroke width, other features and the probability of a CC into a CRF model to determine whether the CC is a character. This procedure is executed parallel on every binary image. At last, text line candidates obtained in every binary image are merged as final result. By this way, text in color images can be detected accurately and comprehensively. In the future, the proposed algorithm can be used in image retrieval/understanding.