Search

Maliciously edited images, also referred to as tampered or forged images, have become ubiquitous and challenging in recent times with the advent of social media platforms. Image splicing is a kind of image tampering where regions of images are copied and pasted into another image. Image tampering detection is a method for the authenticity verification of images produced as evidence in the court of law and forms a sub problem in image forensics. The aim of our work is to detect and locate the boundaries of image splicing, without a-priori knowledge of the reference image. We propose two methods, of which the first uses classification of texture patterns along with the standard deviations of block discrete cosine transformation of textures. In the second method, we combine the first with image quality artifacts incurred due to image tampering along with the entropy of histograms to get an integrated method. We use a support vector machine classifier for training and testing these approaches. Our methods are tested on two publicly available tamper detection datasets and the results are compared with seven other existing works. It is observed that our integrated method gives best results. We have also tested the proposed methods on a social media dataset and the results are encouraging. [ABSTRACT FROM AUTHOR]

Videos are acceptable as evidence in the court of law, provided its authenticity and integrity are scientifically validated. Videos recorded by surveillance systems are susceptible to malicious alterations of visual content by perpetrators locally or remotely. Such malicious alterations of video contents (called video forgeries) are categorized into inter-frame and intra-frame forgeries. In this paper, we propose inter-frame forgery detection techniques using tamper traces from spatio-temporal and compressed domains. Pristine videos containing frames that are recorded during sudden camera zooming event, may get wrongly classified as tampered videos leading to an increase in false positives. To address this issue, we propose a method for zooming detection and it is incorporated in video tampering detection. Frame shuffling detection, which was not explored so far is also addressed in our work. Our method is capable of differentiating various inter-frame tamper events and its localization in the temporal domain. The proposed system is tested on 23,586 videos of which 2346 are pristine and rest of them are candidates of inter-frame forged videos. Experimental results show that we have successfully detected frame shuffling with encouraging accuracy rates. We have achieved improved accuracy on forgery detection in frame insertion, frame deletion and frame duplication. [ABSTRACT FROM AUTHOR]

• We address the problem of thermal to visual person re-identification. • We propose collaborative metric learning method to address the problem. • The proposed collaborative metric learning uses maximum margin matrix factorization. • We obtain the more generalised metric due to the maximum margin. • Proposed method outperforms the baseline algorithms in few-shot learning settings. Thermal to visual person re-identification (T2V-ReID) is a cross-domain image retrieval problem. In this problem, the matching of a person's image takes place, where the image is taken by different cameras (thermal and visual) at different times. This problem has numerous applications in night-time security surveillance. It is challenging due to the large intra-class variations and cross-domain discrepancies. Recently, deep metric learning methods are proposed for this problem. Still, there is a scope to improve the metric learning by generalizing the metric. In this paper, we have proposed the collaborative metric learning using Maximum Margin Matrix Factorization. It uses the group-wise similarities and collaboratively predicts the similarities. We can learn a more generalized metric by utilizing the maximized margin in this method. The proposed method is tested on the RegDB and RGB-D-T data sets, and the method outperforms the existing works in the few-shot learning settings. [ABSTRACT FROM AUTHOR]