Adequate recognition of lips posture for speech articulation analysis requires of the measurement of several anthropometric mouth parameters. These are needed to estimate the position and contour of the lips and teeth and tongue positions, as well. Here, a method is proposed for lips contour detection under natural conditions without any extra hardware requirements for image acquisition. The purpose of the suggested process is to obtain the lips contour based on red hue fields detection. Afterward, geometrical features of lips are extracted from their detected contour. Image processing is divided into the following steps: Face and mouth search, lips contour detection and feature estimation from lips geometry. Due to high dimensionality of the initial feature space, it is very important to evaluate the performance of the lips features regarding their ability to discriminate pathological lip postures in the case of children with cleft lip and palate. In this paper, a proposed method for effective selection of the image feature set was developed using multivariate analysis techniques. Finally, the discriminant performance of the selected training sets was evaluated using bayesian estimators. Results of the comparison for different common testing algorithms show that the proposed processing method exposes better performance.