Facial landmark automatic identification from three dimensional (3D) data by using Hidden Markov Model (HMM)

Landmark identification is one of the fundamental aspects and a necessary step of Three Dimensional (3D) data process. Numerous methods for landmark identification have been proposed, but none with high efficiency, good robustness and strong adaptability has been found yet. In this study, a novel method was developed and applied on a 120 subject's database to automatically identify facial landmarks from 3D facial scanned data. Spin Image (SI) is used to extract local feature and Hidden Markov Model (HMM)is adopted to implement the landmark identification procedure. Eleven HMM5 were trained to identify different facial landmarks, respectively. We developed a three -hierarchy experiment to test the validity and reliability of the algorithm. Results show that SI is highly efficient and robtist to extract facial feature. The maximum landmark recognition rate reached 95.9%. The influence of parameters in SI on the validity and reliability of landmark identification was also investigated. It was found that Bin Size of SI can improve or reduce Identification Accuracy Rate (IAR) by Bin Size value variation. The mean value of IAR increases with the Bin Size until Bin Size reaches 10, when IAR acquires its maximum value 100% and remains constant before Bin Size reaches 65. After that, IAR dropped with the increase of Bin Size, and the velocity of the drop keeps increasing until it reaches the minimum value. in contrast, Support Angle of SI influences IAR positively. Support Angle starts to function at the value of 10. Then, IAR increases with it until Support Angle reaches the degree of 90, when IAR acquired and maintained a constant maximum value of 100%. At last, a comparison was made to address the superiority of HMM over Artificial Neural Network (ANN) on facial landmark identification and we figured out the direction of future work as well. Besides the parameters of SI as discussed in this study, we can investigate the influence of the point sampling granularity on the identification accuracy as well in the future. Relevance to industry: Processing the 3D anthropometric data in order to make it useful for the design of products and facilities is still a challenging problem. One of the typical difficulties is that there is no valid human anatomical landmarks identification method. The improvement of reliability and accuracy of the automatic 3D landmarks identification algorithm for various human body shapes will reduce the time and cost for performing manual landmark palpating and dimension extraction. This work is expected to bring benefits to human -dimension -related ergonomic design and improvement. (C) 2016 Elsevier B.V. All rights reserved.