Criminal Investigation Image Classification Based on Spatial CNN Features and ELM
Daxiang Li, Xin Qiu, Zhiyu Zhu, Ying Liu · 2018
With the advent of the era of big data, the image of the criminal investigation has been explosively growing. To effectively help police officers classify suspect targets, a new classification method of criminal investigation image that combines spatial convolutional neural network (SCNN) and extreme learning machine (ELM) is proposed. Firstly, the image sub-blocks of different space are obtained by a five-block preprocessing on the criminal investigation image, which is part of the original image. Secondly, different image blocks are sent to a fine-tuned convolutional neural network to extract the features of the criminal investigation image. Finally, a criminal investigation image classification algorithm called SCNN-ELM is proposed in combination with extreme learning machine. The experimental results based on 2800 real criminal investigation image database and the standard Corel 1K image database show that the proposed method has an average accuracy improvement of 7.98% compared with the image classification method based on non-blocking CNN features, and the average classification accuracy is also superior than other similar methods.