NEC says "these developments are expected to significantly reduce the time and cost of manual content inspections as well as improve the scale and accuracy of content assessment."
Among the features of the video content identification technology are:
Accurate detection of copied or altered video content
Video signatures are extracted for each frame based on differences in the luminance between sets of sub-regions on a frame that are defined by a variety of locations, sizes, and shapes. Video signatures represent a unique fingerprint that can be individually detected frame by frame. This technology is capable of accurately detecting video content with that was created with such editing operations as analog capturing (*3), re-encoding (*4) and caption overlay (*5), which was conventionally very difficult to detect.
A high detection rate and low false positive rate for all video contents
By estimating confidence of signatures generated from each frame and using the confidence for sequence identification, the technology achieves a high detection rate (*6) with a very low false positive rate (*7). These technologies achieved an average detection rate of 96% at a very low false alarm rate of 5ppm (5 in one million) through tests conducted by the international standardization organization.
Detection of short video scenes
Due to the high identification capability of signatures, the technology is capable of accurately detecting video scenes as short as 2 seconds (60 frames), which was formerly impossible when using conventional methods.
Compatibility with home PCs
By designing a compact signature size of 76 bytes per frame, the storage memory required for the matching process is minimized. As a result, a home-class PC (*8) can match approximately 1,000 hours of video in 1 second. Due to the proliferation of video distribution services on the Internet, the detection and deletion of illegally distributed and copied video content (copyright infringement) has become a crucial issue for content holders and service providers. This problem has conventionally been addressed by manual inspection which is incapable of accurately tracking the constantly growing volume of Internet content. In order to solve this issue, various automatic detection tools have been proposed, such as digital watermarks where content is embedded with special code, and the use of image retrieval technology. However, it was prohibitively difficult to accurately inspect large databases, short content or video produced through various editing operations. These new technologies resolve each of the above issues.
The tech was approved in late April, and will be officially published sometime during September.
NEC plans to demo the tech this week.