Speaker: Prof. Prabin Kumar Bora, Professor, Department of Electrical and Electronics Engineering, Indian Institute of Technology Guwahati, India
Prof. Prabin Kumar Bora Biography: Prof. Prabin Kumar Bora received his M.E. and PhD degrees from Indian Institute of Science, Bangalore, both in Electrical Engineering. Since 1997, he has been associated with the Department of Electrical and Electronics Engineering, Indian Institute of Technology Guwahati where he is a professor now. His area of interest includes image processing and computer vision and the application of signal processing techniques to process physiological and communication signals. He has guided 10 PhD students and over 60 M.Tech. students. He has published more than 120 journal and international conference publications.
Abstract: Illumination has been found to be an effective cue for detecting splicing forgeries in images. In illumination-based image forensics, the illumination information is estimated from different objects present in an image. The estimated illumination features are later compared to check for possible forgeries. The forensics community has been using two different aspects of illumination for forgery detection: the illumination direction and the illumination colour. Hany Farid and his team at Dartmouth College, pioneered the work on illumination direction-based forensics. In their first method, they estimated the 2D illumination direction from the shading and 2D object contour normals and checked the consistencies in the illumination directions. Later, they proposed to use spherical harmonic (SH) analysis to estimate more complex illumination environment in terms of the SH coefficients, making the method more applicable to real-life images. To estimate the full 3D illumination environment, a 3D face model is created from some face images. The 3D surface normals for a test face image are extracted by applying this model and used to estimate the lighting environment. Recently, Peng et al. have proposed a more accurate 3D SH-based method by relaxing some less practical assumptions about human faces. Like the illumination direction, the illumination colour is proved be an effective cue for checking authenticity of images. In the first illumination colour-based method, we showed how illumination colour can be effectively used to expose splicing forgery. In the method, we first created a dichromatic plane (DP) from the specular highlights of each object utilising the dichromatic reflection model (DRM). We showed that for an authentic image, the DPs estimated from different objects intersect at a single point. For a forged image, there will be more than one intersection points. Carvalho et al. from Brazil have proposed a machine learning-based approach, where a new image, called illuminant map (IM), is first created by replacing each homogenous region with its illuminant colour. Then, machine learning-based classifiers are trained to capture the texture, shape and colour inconsistencies present in spliced images. The presentation will outline these techniques.