Enhanced-IPMH as a Robust Visual Descriptor from H.264/AVC and Evaluation of Parameters Effects

Intra-prediction Modes-based (IPM-based) descriptors are among robust and competitive visual descriptors for near-duplicate video similarity detection, in general and content-based copy detection (CCD), in particular. IPM-based descriptors are extracted from the compressed H.264/AVC (MPEG-4/AVC) video domain. Intra-prediction Modes (IPM) are the building blocks of the key frames (I and IDR slices) in the H.264/AVC video standard. IPM-based descriptors are generally constructed based on the probability distribution of the unified intra-prediction modes of the key frames. The current research introduce an enhanced version of IPM-Histogram (IPMH) with 10 bins, which is called enhanced-IPMH (e-IPMH). This research conducted using a subset of TRECVID/CCD (2011), dataset and TREC-EVAL-Video software to compute the performance measures. Based on the experimental evidences, the e-IPMH is an effective and inexpensive visual feature, compared to the pixel domain global descriptors. Analysing the experimental results of the e-IPMH, compared to its predecessor, IPMH shows improvement in the performance measures: Mean Reciprocal Rank (MRR) and Precision@1. However, its mean processing time, reveals it is slower compared to IPMH, due to its larger descriptor size. The current research also conducted a series of experiments to evaluate the effect of spatio-temporal parameters on IPM-based descriptors. The scope of the experiments are limited to the content-preserving visual distortions: T3, T4, T5 and T6 which are the functional scope of global visual descriptors.