Facial Expression Recognition Using Local Positive Directional Pattern (LPDP)
Naga Raju Katta1, M. Babu Reddy2
1Naga Raju Katta, Research Scholar, Department of Computer Science, Krishna University, Machilipatnam (Andhra Pradesh), India.
2M. Babu Reddy, Assisstant Professor and HOD, Department of Computer Science, Krishna University, Machilipatnam, Krishna (Andhra Pradesh), India.
Manuscript received on 21 March 2019 | Revised Manuscript received on 02 April 2019 | Manuscript Published on 18 April 2019 | PP: 61-70 | Volume-7 Issue-6S March 2019 | Retrieval Number: F02150376S19/2019©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: This paper proposes an efficient facial expression system (FES) that can be adapted for many real time applications. Many researchers are concentrated on real time facial expression system, but there are still various issues to be solved like noise due to various reasons. The pixels in real time images are distributed based on the distances with respect to camera. This is also one of the reasons for noise. The performance of FER system is mostly dependent on robust feature extraction. From the literature survey, it is observed that most of the local descriptors are defined and derived on 3×3 neighborhoods and here central pixel is characterized with 9 pixels and these descriptors shown good performance in different applications. This paper proposes new descriptor named as local positive directional pattern (LPDP) which is derived on 5×5 neighborhood which consisting of 25 pixels. Here central pixel is characterized with 25 pixels so that the proposed descriptor has more discriminative power than exiting local descriptors. This descriptor captured more discriminative features from the facial image by using positives directional responses. For achieving better results, the LPDP code histograms is considered as features that are further processed by generalized discriminant analysis (GDA). This GDA is more helpful for distinguishing the expressions as much as possible that in a non-linear space. Extensive experiments on three kinds of datasets (namely JAFFE, CK+ and CASME-II) prove that the proposed method can improve the accuracy. The proposed approach has shown its superiority by achieving mean recognition rate of 95.23 where other state-art-methods could make 90.21 at the best.
Keywords: FER, Positive Direction, GDA, Noise, Expression.
Scope of the Article: Pattern Recognition