Text detection in video sequence using 1-D DCT

In this paper, a new method of detecting text regions from video sequences, without prior knowledge such as font, color, size of characters is proposed. By analyzing the characteristics of text regions in video sequences, we found that text regions include both horizontal and vertical edges simultaneously, while background regions do not. Therefore, In order to detect text regions accurately, the detection of candidate object regions through 1D-DCT and an adaptive thresholding technique are proposed. Experimental results have shown that the proposed method successfully detects text regions regardless of environmental conditions such as low contrast, complex texture, and various colors and sizes.