Recognition of Facial Emotions Relying on Deep Belief Networks and Quantum Particle Swarm Optimization
Kamal A. El Dahshan, Eman K. Elsayed, Ashraf Mohamed Aboshosha, Ebeid A. Ebeid · International journal of intelligent engineering and systems · 2020
Classification of human face images into emotion categories is a hard challenge.Deep learning is mostly effective for this task.In this paper, we propose a facial emotions recognition system based on Deep Belief Networks (DBNs) and Quantum Particle Swarm Optimization (QPSO).The proposed methodology is composed of four phases: first, the input image is preprocessed by cropping the Region Of Interest (ROI) in order to get the desired region and discard non-significant parts.Second, the ROI is divided into many blocks and then the integral image is utilized to determine the superior (most efficient) blocks.Third, image down sampling algorithm is adapted to reduce the size of the new sub image in order to improve the system performance.Fourth, the emotion's class is identified using the DBN.Instead of adapting DBN parameters manually, QPSO is used to automatically optimize DBN parameters' values.The proposed algorithm has been applied to the Japanese female facial expression (JAFFE) and FER-2013 datasets.By applying the proposed algorithm, the computation time is reduced effectively by about 62% on the whole JAFFE and 82% on the whole FER-2013 datasets.Whereas, the accuracy is retained at 97.2% and 68.1% on JAFFE and FER-2013 datasets respectively.