Indexing of video.
Exact/fuzzy search and recognition of video.

The NBITSearch module allows to create indexes, each of which supports an efficient exact and fuzzy search in billions source video objects.

Video objects can be indexed by mapping them to NBIT-primitives.

Source video objects can be represented:

– by video frames (each video frame is a single object),
– by patches of video frames (each patch is a single object),
– by sets of patches of video frames (each such a set is a single object),
– by series of video frames (each such a series is a single object),
– by any sets of video frames (each such a set is a single object).

The indexing and search of video objects in NBITSearch is being implemented just as indexing and search of images.

Examples of results of video objects search obtained with methods built in NBITSearch are drawn below.



Fig.1.

Fig.1: The upper frame was taken (documentary "Lie in the Name of Science") as a pattern for a search in CROSS mode. The pattern was compressed into JPG first. Then it was stretched and damaged in a way shown in the lower line of frames. Search result: 140 frames which are similar to the pattern including pattern original and 4 frames shown in the lower line.



Fig.2.

Fig.2: The upper frame (documentary "Enigmas of Ancient Egypt. 1 Episode – Mysteries of Seven Pyramids") was taken as a pattern for a search in CROSS mode (normal JPG-compression): Search result: 56 frames which are similar to a given including pattern original and 4 frames shown in the lower line.



Fig.3.

Fig.3: A frame was taken (documentary "Lie in the Name of Science") as a pattern for a search in CROSS mode (normal JPG-compression): Search result: 130 frames which are similar to given including pattern original and 4 frames shown in the lower line.



Fig.4.

Fig.4: The upper frame (documentary "Enigmas of Ancient Egypt. 1 Episode – Mysteries of Seven Pyramids") was taken as a pattern for a search in UNDER mode (normal JPG-compression). Search result: 46 frames which are similar to given including pattern original and 4 frames shown in the lower line.



Fig.5.

Fig.5: The upper frame (documentary "2012") was taken as a pattern for a search in OVER mode (normal JPG-compression). Search result: 13 frames which are similar to given frame including pattern original and 4 frames shown in the lower line.

With the use of a single NBIT-index the exact and fuzzy search can be organized with ignoring one or several areas of a pattern (see Fig.6).


Fig.6.


Search Technology developed with support from FASIE
foundation formed by the Government of Russian Federation
Novosib-BIT LLC 2004 - 2017
Patented