DETECTION OF COPY MORE FORGERY IN DIGITAL IMAGES USING AN OPTIMIZED BLOCK BASED APPROACH

Abstract

Technology is growing exponentially and so does the sophistication of various imaging tools that makes it easier for a person to forge digital images. There are several types of forgeries which are done on images out of which Copy-Move is one of the most widely used. In this type of forgery a portion of an image is copied and pasted over to another portion to conceal some information. Several approaches have been developed so far to detect such kind of forgery. In the block based approach the image is divided into overlapping blocks and each block is matched against every other block for similarity. However such a matching would be computationally expensive, thus each block’s dimensionality is reduced by extracting its vital features and using this information for representation of the block. To extract the features we first applied the Laplacian of Gaussian filter on the image to highlight the areas of interest such as edges, blobs, etc. Subsequently we used Discrete Cosine Transform algorithm on each block to generate DCT coefficient matrix for each block. The DCT coefficients matrix is quantized and averaging is used to extract a feature set with reduce dimensionality. Now these feature vectors with respect to each block are stored in a matrix called feature vector matrix. The matrix is then lexicographically sorted so that vectors corresponding to similar blocks come adjacent to each other. The decision of forgery can then be taken if a number of connected/similar blocks are equidistant to each other. The feature extraction is vital as not only does it reduce the computation complexity but choosing the correct set of features makes an approach invariant to various image manipulations such as rotation, translation, scaling,etc.