Video Text Detection Using Multi-Threshold Shot Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Optical Character Recognition (OCR) systems struggle with accurately detecting text in videos due to blurriness, partial obstruction, and variations in text from frame to frame, leading to inaccurate recognition and failure in identifying textual logos.
Innovation Solution
A multi-step technique using two distance thresholds and shot detection to enhance text detection in videos, incorporating a textual-visual scoring system and shot analyzer to extend frame detection based on visual similarity and confidence scores, thereby improving recall and maintaining precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional OCR is used to detect text in video frames, then the system can process videos, but the recognition accuracy deteriorates due to blurriness, partial obstruction, and frame-to-frame variations
Solution Approach 1:
The video processing is segmented into distinct phases: initial frame detection using strict threshold, shot boundary identification, and subsequent frame detection using relaxed threshold. This segmentation allows different detection strategies to be applied at different stages, improving overall accuracy while maintaining consistency.
Solution Approach 2:
The system performs preliminary detection on the initial frame using a first (strict) distance threshold to establish a baseline set of detected frames. This preliminary action provides a reference point for subsequent detection in the same shot, allowing the system to carry forward reliable detections while filtering out false positives.
2Measurement precision
If a strict distance threshold is used to maintain precision, then false positives are reduced, but recall deteriorates by missing partially obscured or blurry text
Solution Approach 1:
The detection threshold is made dynamic rather than static. The system uses a first distance threshold for initial frame detection and a second (relaxed) distance threshold for subsequent frames within the same shot. This dynamic adjustment allows the system to adapt its precision-recall balance based on the detection stage and context.
Solution Approach 2:
Shot boundary detection acts as an intermediary mechanism that bridges the gap between strict and relaxed detection thresholds. By identifying shot boundaries, the system can apply relaxed thresholds within shots (where context is consistent) while maintaining strict thresholds across shot transitions, thereby recovering lost text information without introducing false positives.
3Measurement precision
If the system processes every frame individually with strict criteria, then precision is maintained, but productivity deteriorates due to computational overhead
Solution Approach 1:
The system maintains continuous detection within shots by carrying forward the detected text state from frame to frame. Instead of independently evaluating each frame against strict criteria, the system continuously updates the detection state using relaxed thresholds, thereby maintaining precision while improving processing efficiency through the continuity of useful action across frames.
Solution Approach 2:
The system applies strict detection periodically at shot boundaries rather than continuously at every frame transition. This periodic application of strict criteria at key transition points maintains overall detection accuracy while reducing computational overhead compared to frame-by-frame strict evaluation.
4Device complexity
If the system uses a single distance threshold, then the detection process is simple, but adaptability deteriorates in handling varying text quality across different video conditions
Solution Approach 1:
The system changes the detection parameter (distance threshold) based on the video processing stage and text quality conditions. By switching between a first strict distance threshold and a second relaxed distance threshold, the system adapts to varying text conditions such as blurriness and obstruction while maintaining a relatively simple dual-threshold framework.
Data Source
AI summary
Systems and methods for detecting text in videos. To address problems with conventional Optical Character Recognition (OCR) systems, the present disclosure provides detection of text for improved OCR. Aspects of the present disclosure can, therefore, be utilized to detect a textual logo in videos, including when the text of the textual logo is clearly visible and when the text is inferred. Thus, examples capture appearance time of a textual logo from a video view perspective. Aspects use a multi-threshold pipeline for detecting video frames including the textual logo. A textual-visual scoring system is additionally used to leverage visual aspects of text in logos. A shot detection system is used to detect inferred text beyond a detected video frame. One or more verification models can be further applied.