Density Evaluation based on Convolutional Networks in Rape Images

Abstract

We evaluate the density of rape pictures based on Convolutional Networks, and compare methods via fused features combined with two kinds of regression approaches: Support Vector Regression, SVR, and Lasso Regression. The Convolutional Networks extract the features of rape images through convolutional layers, pooling layers and activation functions, and then, fully connected layers regress the extracted features to the density value. The fused features involve three types of features: image energy, local binary pattern(LBP) features and Gabor wavelets texture features. First, the method extracts the fused features through python scikit-learn packages [1], and then regression model regresses the fused features to the density value by Support Vector Regression [2] [3] or Lasso Regression [4].