Scene Text Extraction Based on HSL

The characteristics of scene text include the uneven lighting and the variability of character color which make scene text extraction more difficult than document image. In this paper, a novel scene text extraction method is proposed based on the independence between hue and lightness. We categorize text region into three types according to the composed color. Different color information is used to segment different text region. The periodicity and instability of hue value in the hue image are considered. A new local adaptive thresholding method is proposed to segment lightness images due to high contrast of scene text. Experiments show that the proposed method can effectively choose hue or lightness information to segment color scene text region, and obtain better results in terms of precision of character recognition.