A scene text detection method based on color clustering of textual pixels is proposed, in which the initial textual regions of original images are extracted by the maximally extremal stable regional algorithm, and the stable textual pixels are screened out by the stroke width transform algorithm and the angle features. The multi-scale color clustering is conducted in these stable textual pixels, which is combined with a support vector machine for the realization of character regional verification. The text line aggregation is finally adopted to achieve the goal of text detection. The tests are conducted on two public datasets of ICDAR2011 and ICDAR2013, and the F-scores of this algorithm are 0.76 and 0.77, respectively. Compared with the existing text detection methods, the proposed method obtains a good text detection performance.