基于SVM信息融合的图像识别与并行实现

提出用支持向量机（SVM）融合三种基于不同特征表示的表情识别方法进行表情识别,即PCA表情表示、SVM表情表示和FLD表情表示。在用SVM进行特征提取时,提出一种高效的方案选择投影轴。在提取各种特征表示后,对每一种表情特征用1阶最近邻分类器进行初步识别,最后用支持向量机融合这些分类结果进行表情的最终识别。并且针对目前还没有硬件实现情况,提出用TI公司的达芬奇系列的DSP芯片构建并行系统来实现SVM融合算法方案,讨论并优化DSP实现算法的过程,通过实验的结果表明,提出的方案是有效的。

Image recognition based on SVM information fusion and DSP parallel realization

The paper proposes a multi-feature and multi-classifier fusion algorithm for image recognition called Support Vector Machine(SVM) information fusion.First,the paper applies three different algorithms to the feature extraction:PCA,SVM,LDA.The outputs of facial expression classifier based on three expression representations are input to SVM information fusion to get facial expression recognition.For hardware realization is few,this paper uses the DAVINCI DSP serial productions of TI company to realize the SVM information fusion algorithm scheme based on parallel structure.It discusses the process of algorithm realization by DSP.The actual results prove that SVM information fusion algorithm for expression recognition is effective and can be realized by DSPs.