Penyaringan dan pemilihan fitur statistik asas untuk pengecaman spesimen forensik balistik
Firearms identification has been getting very important in crime investigation in the last two decades. In this paper, a recognition system for firearms identification based on cartridge case image is introduced. Cartridge case is one of the important clues towards solving the gun file. There are...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Universiti Kebangsaan Malaysia
2009
|
| Online Access: | http://journalarticle.ukm.my/64/1/ |
| Summary: | Firearms identification has been getting very important in crime investigation in the last two decades. In this paper,
a recognition system for firearms identification based on cartridge case image is introduced. Cartridge case is one of
the important clues towards solving the gun file. There are more than 30 marks left on the surface of the cartridge case
when a gun is fired which are invaluable towards identifying the firearm used. These marks in combination produces a “fingerprint” for identification of a firearm. Therefore, the aim of this research work is towards extraction and identification of suitable features for firearms recognition. Firstly the cartridge case images are segmented into three parts, forming three sets of images. These images were also preprocessed to form three more sets of images. Features were extracted from these original and preprosessed. Twenty significant features each were identified and computed for the original and the preprocessed images. All processing were done using MATLAB programming. A scheme based on correlation analysis was introduced towards features selection based on the concept of minimising data redundancy but maximising classes’ differences. Features that are highly correlated were dropped and eventually there are only seven significant features. The seven features formed a feature vector for the fireams recognition and were tested on five pistols
of the same model using discriminant analysis. The classification results show that more than 80% of the cartridge case images were classified correctly. |
|---|