Concatenated edge and co-occurrence feature extracted from Curvelet Transform for human detection

An efficient feature extraction method based on the Curvelet Transform for detecting human in static images is proposed in this paper. The edge features can be extracted with the block-based statistical information of each sub-band coefficients, and then the texture feature can be extracted from the co-occurrence of the lowest sub-band coefficients, all the extracted features are concatenated as the final feature vector of the images. All the training and test data are from the INRIA and MIT human dataset. The classification results with test data show that the proposed feature extraction method is suitable for human detection. From the detection results, it can be seen that, the detection accuracy of the proposed method is higher, meanwhile the false alarms are lower than the Histograms of Oriented Gradients(HOG) method.