Enhanced K-means Clustering Technique based Copy-Move Image Forgery Detection

Abstract

Currently, different methods such as copy-move, image morphing, image splicing, image retouching, etc. are used to alter digital images. Among them, copy-move image forgery is one of the most familiar cyber-crimes and is affecting the world of digital images as this type of forgery is easy to create and one of the hardest forgeries to detect. This paper focusses on the detection of blind copy-move image forgery, which is frequently put into practice in the field of passive forensics to confirm the genuineness and integrity of digital images. In the proposed work, discrete wavelet transform is first applied on the input image. The resulting lowest frequency approximate sub-band is partitioned into small overlapping blocks having fixed size with a sliding factor of one pixel. The 2D discrete cosine transform is then computed for each fixed size block and is then stored as a one-row vector by using the zigzag scanning. The use of the hybrid transform together with fast k-means clustering technique helps to increase the processing speed and reduces the overall forgery detection time. The performance of the propounded system is assessed by using Matlab (R2016a) and then compared to the others works and is found to be satisfactory in terms of precision, recall, F1 Score, and forgery detection time.