一种多尺度小波核极限学习机的图像检索仿真

摘要

阐述了核极限学习机原理。在此基础上提出了一种多尺度小波核极限学习机，将多尺度小波核作为极限学习机的核函数，测试表明是其一种可实现的极限学习机核。同时在无训练数据分布的空间也具备分类能力，同等条件下高斯核极限学习机却不具备分类能力。在图像检索中应用多尺度小波核极限学习机，实验表明，相比支持向量机学习机分类算法，该分类算法可提高检索精度以及速度，具有优良的性能和一定的应用价值。%The principle of kernel Extreme Learning Machine (ELM) is demonstrated. Based on above, a multi-scale wavelet kernel ELM is proposed in which the multi-scale wavelet kernel is employed as kernel function. It is an achievable ELM which has classification ability in the distribution space with no training data while Gaussian kernel ELM does not perform well. Simulation results show that multi-scale wavelet kernel ELM has higher retrieval precision and efficiency compared with support vector machine model when they are used in image retrieval. The proposed approach has excellent performance and application value.