Video Layout Area Detection Using Saliency Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing layout detection techniques in video images are inadequate for scenes with little motion, require analysis of multiple consecutive images, or rely on prior knowledge of logos, making them inefficient and limited in adaptability.
Innovation Solution
The method employs a saliency map generated from a visual attention model to detect layout areas by searching for maximum salience points in predefined areas of the image, particularly in corners, and uses criteria to select relevant areas, thereby generating a reduced size image that excludes non-interesting layout areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image to image difference technique is used to detect layouts, then layout detection can be performed, but it fails for scenes with little or no motion and requires analysis of several consecutive images
Solution Approach 1:
The patent replaces the mechanical motion-based detection approach (image to image difference) with a visual attention model that simulates human eye behavior. This model uses saliency maps to identify important regions based on visual features like contrast, color, and texture, rather than relying on motion between frames. This substitution enables reliable layout detection in both dynamic and static scenes.
Solution Approach 2:
The patent changes the detection parameters from motion-based metrics (pixel differences between frames) to visual attention-based metrics (saliency values). By computing saliency maps that highlight regions of visual interest based on local contrast, color distribution, and texture variations, the system can detect layouts in static scenes where motion-based methods fail.
2Reliability
If visual content comparison with predefined layout base is used, then layout detection can be performed, but only layouts contained in the base can be detected requiring prior knowledge
Solution Approach 1:
The patent enables the system to detect layouts autonomously without requiring a predefined database of layout templates. The visual attention model automatically identifies salient regions in the image that correspond to layouts based on their visual characteristics (high contrast, distinct color patterns, textural features). This self-service approach allows detection of both known and unknown layouts, eliminating the need for prior knowledge or training data.
Data Source
Figure 1
Figure 2
Figure 3~4B
AI summary
The invention relates to an automatic detection method in a source image, of at least one area called a layout area comprising at least one layout, such as a logo and/or a score. According to the invention, the layout areas of a source image are detected using the salience of source image pixels. The detection is carried out in specific areas of the source image saliency map, usually in the areas corresponding to the corners of the image or to the bands in the upper part and lower part of the image. In these areas, two points are sought having maximum salience values and distant by at least p points from each other. These two points corresponding to the beginning and end of a layout area. The window bounding these two points then corresponds to a layout area.