This paper proposes a scene text detec tion method based on color distance minimi zation and maximum color difference (MCD).Firstly,multip le of K-means clustering and color distance minimization are performed ,which can extract text connected components from images with different complexities.Considering color clustering method is usuall y influenced by luminance,we use maximum color distance to extract text connect components as supplements,which combines color with gradient information and can overcome the luminance influence to some extent.Then,the obtained connected components are grouped to form text lines following some character merging rules.The candidate text lines usually contain some falsely detected no n-text ones.To improve text detection accuracy, feature extraction and machine learning algorithms are finally used to identify the obtained candidate text lines,and we can get text localization results at last.The proposed method is evaluated on some publicly available datasets,including the ICDAR2011and ICDAR2013datasets.For the ICDAR2011dataset,the achieved recall,precision an d f-measure are 0.66,0.77and 0.71,respectively, and those on the ICDAR2013dataset are 0.65,0.77and 0.70,respectively.The co mparative results with some other text detection methods verify the feasibility and effectiveness of our method.