Towards robust skin colour detection and tracking

A discriminative fusion framework for skin detection

“…Moreover, each detector can vary from a simple classification rule to complicated skin detection algorithms. Two simple skin RGB space detectors are used in this paper as D (1) and D (2) according to the following rules: D (1) : A pixel with value (r, g, b) is skin if r > 95 D (2) : A pixel with value (r, g, b) is skin if |r−g| > 15 To evaluate the proposed framework, our model is trained using just one sample image and its ground-truth. Thereafter, we applied our method to eight test images with different conditions to some extent.…”

“…The ground-truth label images are shown in second column. The third and the forth columns consist of detection results correspond to D (1) and D (2) . The detections of our proposed framework are represented in the last column.…”

“…To quantitatively evaluate the methods, we compare the D (1) , D (2) and the proposed framework by the means of Precision, Recall and F-measure criteria which are defined …”

“…Given the test X and the decisions of the detectors, D's, the task of assigning labels to pixels of X is called inference, where the goal of inference is to find the maximum a posteriori (MAP) solution Y * which is our desired problem in (2). In this stage, an algorithm (loopy belief propagation [12] in this paper) attempts to maximize (3) given the observation X and the tuned parameters λ.…”

“…Detecting skin in images is considered as one of the important preliminaries for computer vision tasks such as face detection, face recognition, and video surveillance [1], [2].…”