Abstract: In order to improve the accuracy of highway pavement crack detection， a new type of surface defects detection algorithm was proposed based on subregion and multiscale analysis. Gray， entropy and texture features distribution information were extracted in cracks and the surrounding areas from different scales of images. Feature vectors of parameters containing the direction trend and bending degree were acquired， and the crack location was detected through learning the support vector machine (SVM) and judging the eigenvector. Experimental results demonstrated that the resistance to noise， versatility and detection accuracy were improved effectively by the proposed algorithm in comparison to the other pavement cracks detection algorithms. The ideal crack detection effect was achieved， and the requirements of highway quality inspection were met effectively.

Key words: crack detection, subregion, multiscale, direction trend, bending degree