Enhancing Facial Feature Detection: Hybrid Active Shape and Active Appearance Model (HASAAM)

Face landmarking is a primary goal of many projects that lead to face-preparing activities, like biometric recognition and mental state comprehension. Despite the inherent diversity of faces, the topic has proven to be extremely difficult due to a wide range of perplexing variables, such as position, expression, illumination, and occlusions. The aim of the HASAAM integrated fitting model is to find new solutions for the feature identification issue by combining the strengths of the Active Shape Model (ASM) and Active Appearance Model (AAM) to provide unique findings on the feature detection problem. In the first step, the ASM was used to identify the external shape landmarks of the face, and in the second stage, the AAM was used to identify the interior form landmarks. Then, the two kinds of landmarks are combined to generate the full face landmark. One issue with ASM is that it can’t produce an optimal global result since it will unavoidably converge to neighborhood minima. Nonetheless, ASM produces precise fitting results for the face’s exterior features, which need essential gradient values. In order to prevent AAM from fitting in order to address the local minima problem, ASM was used to identify these exterior landmarks of the face. The trials were run using the MORPH and LFPW datasets, which are available to the general public. In comparison to other techniques like ASM and AAM, the proposed hybrid model experiment result demonstrated effectiveness in extracting facial features with error rates of 2.2628% for the LFPW database and 2.7174% for the MORPH database. This better result helps to compensate for variations in shape and texture. The suggested hybrid method may be expanded to recognize gender, estimate head posture, estimate gaze, find facial feature points in the presence of significant pose fluctuations, and detect facial expressions.