Quantitative Inspection And Identification Of Striated Tool Marks

Inspection and identification of tool marks play an important role in the investigation and litigation of criminal cases.By analyzing the similarity between the marks extracted from the scene and the resemblance to the marks reproduced in the laboratory,the inspectors determine the specific tool.The prerequisite for this work is that tool marks should be unique and reproducible.In fact,this is the theoretical basis for tool marks inspection and identification.This thesis presents two quantitative statistical identification methods for tool marks based on features of 1000 scratch marks made by 10 unused screwdrivers,1000 cutting marks made by 10 pliers and 800 cut marks made by 8 bolt cutters.The experimental results show that as long as the location of the tool location is accurate,the method of this paper has a strong ability to identify the scratch marks of the screwdrivers and the cutting marks of the clamps.Besides,the two methods accept 2D tool mark image as input and have good illumination invariance.The first method uses four LBP derivatives operators to extract the effective information from the 2D mark images of the three tools and generate the feature vectors.Then,the random forest algorithm is used to supervise and train the feature vectors with the labels.The experimental results show that the trained RF model has a high recognition rate for the test samples and can achieve the identification effect.The second method is a feature learning method based on Deep Convolutional Neural Network(DCNN)with the transfer learning theory.Firstly,the VGG16 pre-trained convolution network layer is used to extract the effective features of the 2D trace image,and then replace the original full connection layer with the new full connection layers,and train the new full connection layers by minimizing the loss function.Similarly,the fine tuned neural network model can perform well the identification of tool mark images.The two methods proposed in this paper are simple and effective,and have some practical application value.At the same time,the higher accuracy of identification also provides strong support for the theory of identification and reproducibility of tool marks.