Content-based obscene video recognition by combining 3D spatiotemporal and motion-based features

Abstract

In this article, a new method for the recognition of obscene video contents is presented. In the proposed algorithm, different episodes of a video file starting by key frames are classified independently by using the proposed features. We present three novel sets of features for the classification of video episodes, including (1) features based on the information of single video frames, (2) features based on 3D spatiotemporal volume (STV), and (3) features based on motion and periodicity characteristics. Furthermore, we propose the connected components' relation tree to find the spatiotemporal relationship between the connected components in consecutive frames for suitable features extraction. To divide an input video into video episodes, a new key frame extraction algorithm is utilized, which combines color histogram of the frames with the entropy of motion vectors. We compare the results of the proposed algorithm with those of other methods. The results reveal that the proposed algorithm increases the recognition rate by more than 9.34% in comparison with existing methods.

Learning Strategies for Sensitive Content Detection

“…The experiments were carried out on 1500 video segments (500 adult videos and 1000 documentaries), while the test set consisted of 18,313 scenes (1103 scenes with at least one sample of sensitive content). The results obtained were better than the work of [89,90], being slightly less efficient in detecting pornographic material that contains a tiny portion of skin colour in each frame. In addition, this approach performs better in the absence of skin colour, unlike methods that do not use skin colour as a part of the classification, as described in [87].…”

“…Behrad et al [90] proposed to identify the largest section of skin colour by extracting six motion-based features employing the Fourier transform of the inter-frame autocorrelation. The classification of pornographic videos is performed using an SVM classifier, obtaining an accuracy of 95.44% over 2000 episodes of pornographic videos and 2000 episodes of conventional videos.…”

A generalized model for relative phases based on bilinear representation of natural image series

“…Now, the localized, orientational basis is widely used in various researches including learning higher level structures and many recognition applications. As an important topic, how to capture the abstract property of motion from statistical natural videos based on these representations draws many attentions both in neuroscience and computer science .…”