Logos extraction on picture documents using shape and color density

Logos detection on textual images is a decisive stage in documents recognition and classification system. The over or the sub-detection of logos strongly penalizes the system capacities and corrupts the subsequent stages result. We developed here an effective and robust logo extraction algorithm while considering the two principals proprieties of logos: spatial compactness and colorimetric uniformity. First, the image content is reduced and transformed using mathematical morphology operators to decrease the distance between the identical logo parts. Afterwards the logo regions of height spatial and chromatic densities are detected. The results demonstrate the robustness of the proposed method over a range of representative text images.