Automatic Actor Recognition for Video Services on Mobile Devices
Face recognition is one of the most promising and successful applications of image analysis and understanding. Applications include biometrics identification, gaze estimation, emotion recognition, human computer interface, among others. A closed system trained to recognize only a predetermined number of faces will become obsolete very easily. In this paper, we describe a demo that we have developed using face detection and recognition algorithms for recognizing actors/actresses in movies. The demo runs on a Samsung tablet to recognize actors/actresses in the video. We also present our proposed method that allows user to interact with the system during training while watching video. New faces are tracked and trained into new face classifiers as video is continuously playing and the face database is updated dynamically.
Analytics-Modulated Coding of Surveillance Video
Video surveillance systems increasingly use H.264 coding to achieve 24 x 7 recording and streaming. However, with the proliferation of security cameras, and the need to store several months of video, bandwidth and storage costs can be significant. We propose a new compression technique to significantly improve the coding efficiency of H.264 for surveillance video. Video content is analyzed and video semantics are extracted using video analytics algorithms such as segmentation, classification and tracking. In contrast to existing approaches, our Analytics-Modulated Compression (AMC) scheme does not require coding of object shape information and produces bit-streams that are standards-compliant and not limited to specific H.264 profiles. Extensive experiments were conducted involving real surveillance scenes. Results show that our technique achieves compression gains of 67% over JM. We also introduced AMC Rate Control (AMC RC) which allocates bits in response to scene dynamics. AMC RC is shown to significantly reduce artifacts in constant-bitrate video at low bitrates.
Adapting video delivery based on motion triggered visual attention
Cues from human visual system (HVS) can be used for further optimization of compression in modern hybrid video coding platforms. We present work that explores and exploits motion related attentional limitations. Algorithms for exploiting motion triggered attention were developed and compared with MPEG AVC/H.264 encoder with various settings for different bitrate levels. For the sequences with high motion activity our algorithm provides up to 8% bitrate savings.