Unsupervised-Image-Forgery-Detection-Using-Discrete-Cosine-Transform-with-K-means-clustering

To detect the forgery in digital images the steps in our method are as follows : (1) first of all, if the image is (RGB) then we are converting the image to gray-scale, (2) In next step, we are dividing the converted image to many overlapping blocks, in our method the size of these blocks is taken as 10x10, (3) Now, we are applying the DCT on all the blocks in order to extract the features from the blocks and storing them in a matrix, (4) now clustering all the blocks using the k-means algorithm in order to classify or differentiate the extracted features from different blocks, (5) in last step, some lexicographical sorting (here, radix sort) is done on the clusters to sort them in order of most matching to least matching. Finally, In the end, results produced are shown in terms of different parameters such as: (i) type of morphological operation on image, (ii) number of clusters used, (iii) correlation limit, (iv) distance limit.