Automatic text detection in complex color image

Detection of text in color images of complex colored background is a very challenging problem. In this paper, an efficient automatic multi-feature fusing text detection method is proposed. First, we generate candidate text regions by merging bounding blocks, which are extracted using the color feature in the spatial color quantized map and the edge feature in the edge map obtained by Sobel's operators followed by mathematical morphology operators. The corner feature and heuristics based edge and color features are then used to eliminate the false candidates. Experimental results on book covers and natural scene images show that this method can detect text including Chinese and English characters accurately. In addition, this method is expected to be more accurate for Chinese character detection, because the verification uses more prior knowledge aimed at Chinese characters.