LoG Feature Extraction Based Photographic Detection

Nowadays powerful digital image editing software makes image modification possible. There are many image processing and computer graphics techniques to manipulate images. This undetermines our trust in photographs. So a forgery detection method is proposed that uses inconsistencies in the color of illumination. This method is applicable to images containing two or more persons and no human interaction for tampering decision. For this, first compute the illuminant estimate of similar regions of the image. Then from the illuminant estimate textural and gradient features are extracted which are given to a machine learning approach for decision-making automatically. The illuminant color is estimated using a statistical gray world method and for extracting textural features, SASI is used. The gradient features are extracted using LoG feature extraction method. Log focuses on regions of high intensity change and it gives better result than any other feature extraction algorithms. These features are combined using a machine learning approach. For this, a SVM classifier is used. Thus photographic forgery can be detected.