基于线性解码和深度回归预测的图像分类算法

IMAGE CLASSIFICATION BASED ON LINEAR DECODING AND DEEP REGRESSION PREDICTION

针对图像分类研究中的分类器输入范围限制和缩放问题,提出一种基于线性特征解码和深度回归模型图像分类算法。首先,通过线性解码器学习低分辨率图像的特征参数;然后,运用学习到的特征对原始高分辨率图像进行卷积和池化操作,得到特征矩阵;再通过Softmax回归模型对图像进行深度学习和分类;最后用距离度量算法得到图像分类结果。实验结果从多方面对比和验证了该方法在分类效率方面超越了传统的误差反向传播算法BP和K最近邻分类算法KNN。

Aiming at the problems of the limitation of classifier input range and the zooming in image classification research,this paper proposes an image classification algorithm which is based on linear feature decoding and deep regression model. First,we learn feature parameters of low-resolution image through linear decoder; secondly,by using the learned features we convolve and pool the primitive high-resolution image to obtain the feature matrix; thirdly,by using Softmax regression model we carry out deep learning and classification on image; finally,we obtain the classification results with distance metric algorithm. Experimental results contrast in many ways as well as verify that our method is superior to traditional error back propagation algorithm and K-nearest neighbour classification algorithm in classification efficiency.