基于小波池的图像分类算法研究

Image Classification Algorithm Based on Wavelet Pooling

本文提出了一种基于二级小波分解的小波池算法。算法放弃了最近邻插值方法,取而代之的是一种有机的、子带的方法,方法能够更准确地表示特征内容,并且伪影更少。将所提出的池化方法与最大池化、平均池化、混合池化、随机池化进行比较,在基准图像数据集MNIST和CIFAR10上测试此方法,证明此方法能产生相等或更优结果,有很好的有效性。

This paper proposes a wavelet pool algorithm based on two-level wavelet decomposition.The method abandons the nearest neighbor interpolation method and replaces it with an organic,sub-band method,which can more accurately represent feature content and have fewer artifacts.It compares the proposed pooling method with maximum pooling,average pooling,mixed pooling,random pooling,and tests these methods on the benchmark image datasets MNIST and CIFAR10 to verify its effectiveness and the ability to produce near-equal or better results.