高额的存储与计算成本限制了神经网络模型在低算力平台的应用,为提高神经网络模型的实用性,提出了两种组合优化算法,通过对多个轻型并行神经网络在连续时间窗口内的概率优化,在保证识别准确率的前提下显著降低了计算成本。为验证算法的可行性,以痛苦表情识别为对象,展开了系列对比实验。实验表明在保持相似准确率的前提下,其计算量相比传统深度学习算法极大降低,提高了神经网络的实用性,并极大降低了存储与计算成本。
The high storage and computing cost limits the application of neural network model in low computing power platform. To overcome the above shortages, we proposed two combined probability optimization algorithms to combine multiple light neural networks in a continuous time window, which can significantly reduce computing load under similar accuracy. The proposed scheme gives a general way for related algorithm on embedded hardware. To verify its effectiveness, a group of experiment was executed on the pain expression recognition with continuous peak values. In comparison to other conventional algorithms, the model complexity, computing load and storage of proposed scheme decrease obviously under consistent accuracy.
[1] LEE J, WONG A. TimeConvNets: a deep time windowed convolution neural network design for real-Time video facial expression recognition[C]// 2020 17th Conference on Computer and Robot Vision (CRV). Los Alamitos: IEEE Computer Society, 2020: 9-16.
[2] MICHEL P, KALIOUBY R E. Real time facial expression recognition in video using support vector machines[C]// Fifth International Conference on Multimodal Interfaces. New York: Association for Computing Machinery, 2003: 258-264.
[3] BARGAL S A, BARSOUM E, FERRER C C, et al. Emotion recognition in the wild from videos using images[C]// Proceedings of the 18th ACM International Conference on Multimodal Interaction. New York: Association for Computing Machinery, 2016: 433-436.
[4] ZHSNG T, ZHENG W, CUI Z, et al. Spatial-Temporal recurrent neural network for emotion recognition[J]. IEEE Transactions on Cybernetics, 2019, 49(3): 839-847.
[5] AN F, LIU Z. Facial expression recognition algorithm based on parameter adaptive initialization of CNN and LSTM [J]. The Visual Computer, 2019, 36(3): 483-498.
[6] HASANI B, MAHOOR M H. Facial expression recognition using enhanced deep 3D convolutional neural networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE, 2017: 2278-2288.
[7] RAJAN S, POONGODI C, Devaraj S, et al. A novel deep learning model for facial expression recognition based on maximum boosted CNN and LSTM [J]. IET Image Processing, 2020, 14(7): 1373-1381.
[8] EKMAN P. Constants across cultures in the face and emotion [J]. J Pers Soc Psychol, 1971, 17(2): 124-129.
[9] MEHRABIAN A. Communication without words [J]. University of East London, 1968, 24(4): 1084-5.
[10] WANG S, ECCKESTON C, KEOGH E. The time course of facial expression recognition using spatial frequency information: comparing pain and core emotions [J]. Journal of Pain, 2020, 22(2): 192-208.
[11] REICHERTS P, WIESER M J, GERDES A B M, et al. Electrocortical evidence for preferential processing of dynamic pain expressions compared to other emotional expressions [J]. Pain, 2012, 153(9): 1959-1964.
[12] LUCEY P, COHN J F, KANADE T, et al. The extended cohn-kanade dataset (CK+): a complete dataset for action unit and emotion-specified expression[C]//2010 IEEE Conference on Computer Vision and Pattern Recognition Workshops. Los Alamitos: IEEE Computer Society, 2010: 94-101.
[13] LYONS M, AKAMATSU S, KAMACHI M, et al. Coding Facial Expressions with Gabor Wavelets[C]//Third IEEE International Conference on Automatic Face and Gesture Recognition. Los Alamitos: IEEE Computer Society, 1998: 200-205.
[14] ABHINAV D, ROLAND G, SIMON L, et al. Collecting Large, Richly Annotated Facial-Expression Databases from Movies[J]. IEEE Multimedia, 2012, 19(3): 34-41.
[15] VIOLA P, JONES M J. Rapid object detection using a boosted cascade of simple features[C]// Computer Vision and Pattern Recognition. Los Alamitos: IEEE Computer Society, 2001: 511-518.
[16] VIOLA P, JONES M J. Robust real-Time face detection [J]. International Journal of Computer Vision, 2004, 57(2): 137-154.
[17] 泰勒, 佩普劳, 希尔斯,等. 社会心理学[M]. 第10版.北京: 北京大学出版社, 2004: 89-90.
[18] HUANG T S. A fast two-Dimensional median filtering algorithm [J]. IEEE Trans on Acoustic Speech & Signal Processing, 1979, 27(1): 13-18.
[19] ARRIAGA O, VALDENEGRO-TORO M, PLÖGER P. Real-time convolutional neural networks for emotion and gender classification[DB/OL]. [2019-12-16]. https://arxiv.org/pdf/1710.07557.pdf.
[20] MA N, ZHANG X, ZHENG H T, et al. ShuffleNet V2: Practical guidelines for efficient CNN architecture design[C]// European Conference on Computer Vision. Berlin: Springer-Verlag Berlin, 2018: 122-138.
[21] SANDLER M, HOWARD A, ZHU M, et al. MobileNet V2: Inverted residuals and linear bottlenecks [C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 4510-4520.
