Splicing detection in color image based on deep learning of wavelet decomposed image
Ali Retha Hasoon Khayeat, Ahmed Abdulhadi Al-Moadhen, Mustafa Ridha Hassoon Khayyat · AIP conference proceedings · 2020
Image splicing is done by duplicate part(s) of an image and pasted in a different image. This basic technique is very common in image forgery therefore it reduced the user confidence in the digital image. The need for a reliable and effective method to detect this type of counterfeiting has been increased. Splicing detection in colour image based on Deep Learning of Haar wavelet decomposed image is developed in this work. The colour image is converted to grayscale and decomposed into 1st level (LL1, LH1, HL1, and HH1). Then to detect splicing, we used the semantic segmentation of the decomposed image. Consider that the Convolutional Neural Network model was used to segment the decomposed LL1 image. The SegNet is applied to boost the semantic segmentation and utilize the classification in our proposed approach. The experimental work confirmed the efficiency of the suggested method where the forgery (splicing) detected in 89% of the tested images with a very high percentage of localization.