Video Stream Indexing via Frame Cross Entropy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video stream alignment techniques are computationally expensive and time-consuming, making them impractical for many applications, especially when dealing with video degradations such as compression, blurring, and affine transformations.
Innovation Solution
A method for indexing a video stream that determines salient points, computes cross entropy values, and sums them to form a frame information number, creating a sequence of index values that facilitates efficient alignment and comparison of video streams, even under various degradations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional techniques such as SIFT, SURF, DAISY, and Harris corner processing are used to identify salient points and perform frame-by-frame comparison, then alignment accuracy is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent extracts only the most essential features for alignment by computing a single information number per frame based on pixel intensity statistics, rather than extracting multiple complex features like SIFT or SURF descriptors. This extraction of minimal necessary information reduces computational complexity while maintaining alignment accuracy.
Solution Approach 2:
The patent segments the video alignment problem into independent frame-level information number computations, where each frame is processed separately to generate an information number, and then these numbers are compared across frames to achieve alignment. This segmentation avoids the computationally expensive frame-by-frame pixel-level comparison of conventional methods.
2Measurement precision
If conventional frame-by-frame comparison techniques are used, then alignment precision is improved, but processing time increases significantly
Solution Approach 1:
The patent performs preliminary action by pre-computing information numbers for each frame based on simple pixel intensity statistics before any alignment comparison is needed. These pre-computed information numbers can then be quickly compared to determine alignment, avoiding the need for time-consuming frame-by-frame pixel comparison during the actual alignment process.
3Reliability
If robust alignment techniques that handle video degradations are implemented, then reliability of alignment is improved, but computational cost increases
Solution Approach 1:
The patent uses parameter changes by computing information numbers based on pixel intensity statistics that are inherently robust to various degradations. The information number captures essential frame characteristics through statistical parameters (mean, variance, skewness, kurtosis) that remain stable under compression, blurring, and other common video degradations, providing robust alignment without additional computational overhead.
Data Source
AI summary
Embodiments including a method and apparatus for indexing a video stream are disclosed. In one embodiment, a method for indexing a video stream comprises accessing a video stream comprising a plurality of frames. For each frame, the method determines salient points computes a cross entropy value for each salient point, and sums the cross entropy values to form a frame information number. A sequence of frame information numbers for the plurality of frames in the video streams forms an index value for the video stream.