基于低秩稀疏的人臉表情識(shí)別方法研究
發(fā)布時(shí)間：2018-08-28 13:31
【摘要】：人臉表情是日常交流的主要方式之一,相對(duì)其他表達(dá)方式而言,能更有效的體現(xiàn)彼此的內(nèi)心活動(dòng)。人臉表情識(shí)別涵蓋了心理學(xué)、生理學(xué)、圖像處理、模式識(shí)別等多個(gè)領(lǐng)域,是一個(gè)交叉性的學(xué)科。在人機(jī)交互領(lǐng)域有著廣泛的運(yùn)用,但由于相關(guān)的識(shí)別技術(shù)還不成熟,在日常的生活運(yùn)用還處在一個(gè)嘗試性的階段,存在識(shí)別率不高等問(wèn)題,因此值得進(jìn)一步深入研究。本文主要的研究?jī)?nèi)容如下:1.針對(duì)協(xié)作低秩分層稀疏表情識(shí)別算法采用隨機(jī)取樣的方式構(gòu)建表情字典,導(dǎo)致表情識(shí)別效果并不穩(wěn)定。因此可以通過(guò)結(jié)合LC-KSVD(Label consist K-SVD)字典學(xué)習(xí)方法,提高協(xié)作低秩分層稀疏表情識(shí)別算法的穩(wěn)定性和準(zhǔn)確度。2.由于LC-KSVD算法在訓(xùn)練字典的時(shí)候,受到最后一次訓(xùn)練樣本的影響更大,并且字典原子間極有可能存在較大相關(guān)性,特別是當(dāng)字典規(guī)模較小時(shí),不能學(xué)習(xí)出有效字典,影響著識(shí)別的準(zhǔn)確度。但如果當(dāng)字典規(guī)模較大時(shí),算法成本又將加大。因此需要設(shè)計(jì)出一個(gè)尺度自適應(yīng)的,且各原子間相干性最低的字典學(xué)習(xí)算法,使得字典能夠以最合適的字典規(guī)模,包含更有效的分類信息。3.基于低秩稀疏的人臉表情識(shí)別方法一般通過(guò)有效的分離表情變化稀疏矩陣,然后在特定表情字典上對(duì)該稀疏矩陣進(jìn)行稀疏表示,以達(dá)到最佳的識(shí)別效果。但實(shí)際運(yùn)用中往往受到一些噪聲干擾,使得相應(yīng)的低秩稀疏分解算法不能有效的分離表情變化稀疏部分,使得低秩稀疏分解算法在實(shí)際的人臉表情識(shí)別運(yùn)用中存在不少的缺陷。因此通過(guò)添加相關(guān)約束項(xiàng),可以將復(fù)雜噪聲從表情序列中分離,并有效的提取表情變化特征,從而提高識(shí)別效率。
[Abstract]:Facial expression is one of the main ways of daily communication. Compared with other expressions, facial expression can more effectively reflect each other's inner activities. Facial expression recognition covers many fields, such as psychology, physiology, image processing, pattern recognition and so on. It is widely used in the field of human-computer interaction, but because the related recognition technology is not mature, the daily life of the application is still in a trial stage, there are problems such as low recognition rate, so it is worth further study. The main contents of this paper are as follows: 1. An expression dictionary is constructed by random sampling for collaborative low rank hierarchical sparse expression recognition algorithm, which results in unstable performance of expression recognition. Therefore, we can improve the stability and accuracy of the collaborative low rank hierarchical sparse expression recognition algorithm by combining the LC-KSVD (Label consist K-SVD) dictionary learning method. Because the LC-KSVD algorithm is more affected by the last training sample when training the dictionary, and the dictionary atoms are likely to have a greater correlation, especially when the dictionary size is small, it can not learn an effective dictionary. It affects the accuracy of recognition. But if the dictionary is large, the cost of the algorithm will increase. Therefore, it is necessary to design an adaptive dictionary learning algorithm with the lowest coherence among atoms, so that the dictionary can contain more effective classification information with the most appropriate dictionary size. The low rank sparse facial expression recognition method usually separates the sparse matrix of expression change effectively and then sparse represents the sparse matrix in a specific expression dictionary in order to achieve the best recognition effect. However, in the practical application, some noises often interfere, which makes the corresponding low-rank sparse decomposition algorithm can not effectively separate the sparse parts of facial expression changes, which makes the low-rank sparse decomposition algorithm have many defects in the actual application of facial expression recognition. Therefore, the complex noise can be separated from the expression sequence by adding correlation constraints, and the feature of facial expression change can be extracted effectively, so as to improve the recognition efficiency.
【學(xué)位授予單位】：昆明理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41
本文編號(hào)：2209532
[Abstract]:Facial expression is one of the main ways of daily communication. Compared with other expressions, facial expression can more effectively reflect each other's inner activities. Facial expression recognition covers many fields, such as psychology, physiology, image processing, pattern recognition and so on. It is widely used in the field of human-computer interaction, but because the related recognition technology is not mature, the daily life of the application is still in a trial stage, there are problems such as low recognition rate, so it is worth further study. The main contents of this paper are as follows: 1. An expression dictionary is constructed by random sampling for collaborative low rank hierarchical sparse expression recognition algorithm, which results in unstable performance of expression recognition. Therefore, we can improve the stability and accuracy of the collaborative low rank hierarchical sparse expression recognition algorithm by combining the LC-KSVD (Label consist K-SVD) dictionary learning method. Because the LC-KSVD algorithm is more affected by the last training sample when training the dictionary, and the dictionary atoms are likely to have a greater correlation, especially when the dictionary size is small, it can not learn an effective dictionary. It affects the accuracy of recognition. But if the dictionary is large, the cost of the algorithm will increase. Therefore, it is necessary to design an adaptive dictionary learning algorithm with the lowest coherence among atoms, so that the dictionary can contain more effective classification information with the most appropriate dictionary size. The low rank sparse facial expression recognition method usually separates the sparse matrix of expression change effectively and then sparse represents the sparse matrix in a specific expression dictionary in order to achieve the best recognition effect. However, in the practical application, some noises often interfere, which makes the corresponding low-rank sparse decomposition algorithm can not effectively separate the sparse parts of facial expression changes, which makes the low-rank sparse decomposition algorithm have many defects in the actual application of facial expression recognition. Therefore, the complex noise can be separated from the expression sequence by adding correlation constraints, and the feature of facial expression change can be extracted effectively, so as to improve the recognition efficiency.
【學(xué)位授予單位】：昆明理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41
【參考文獻(xiàn)】
相關(guān)期刊論文 前6條
1 楊凡;張磊;;基于Gabor參數(shù)矩陣與改進(jìn)Adaboost的人臉表情識(shí)別[J];計(jì)算機(jī)應(yīng)用;2014年04期
2 唐恒亮;孫艷豐;朱杰;趙明茹;;融合LBP和局部稀疏表示的人臉表情識(shí)別[J];計(jì)算機(jī)工程與應(yīng)用;2014年15期
3 周曉彥;鄭文明;辛明海;;基于稀疏表示的KCCA方法及在表情識(shí)別中的應(yīng)用[J];模式識(shí)別與人工智能;2013年07期
4 王志良,陳鋒軍,薛為民;人臉表情識(shí)別方法綜述[J];計(jì)算機(jī)應(yīng)用與軟件;2003年12期
5 金輝,高文;人臉面部混合表情識(shí)別系統(tǒng)[J];計(jì)算機(jī)學(xué)報(bào);2000年06期
6 高文，金輝;面部表情圖像的分析與識(shí)別[J];計(jì)算機(jī)學(xué)報(bào);1997年09期
相關(guān)博士學(xué)位論文 前1條
1 程廣濤;基于壓縮感知的人臉識(shí)別方法研究[D];天津大學(xué);2015年
，本文編號(hào)：2209532
本文鏈接：http://www.sikaile.net/kejilunwen/ruanjiangongchenglunwen/2209532.html
最近更新
教材專著
