Image Summarization Device for Capsule Endoscopy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image summarization methods fail to efficiently summarize large numbers of temporally or spatially continuous images, such as those captured by capsule endoscopes, leading to missed important areas like lesions due to unnecessary image deletion, as they do not adequately consider the relationship between deleted and retained images.
Innovation Solution
An image processing device that performs a dual-image summarization process, combining similarity-based image reduction and target object/scene recognition to determine which images to delete, ensuring that critical areas are preserved by calculating deformation information and coverage ratios to control the observation of non-covered areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If images are deleted from the original image sequence to summarize it, then the number of images is reduced and checking time is decreased, but important areas such as lesions may be missed
Solution Approach 1:
The patent segments the image sequence into multiple groups based on similarity, and performs summarization at the group level rather than individual image level. This allows systematic control over which images are deleted while ensuring representative images from each group are retained, thus reducing checking time while maintaining disease detection reliability.
Solution Approach 2:
The patent performs preliminary analysis to calculate similarity between images and identify representative images before the actual summarization process. By pre-identifying which images contain important information and which are redundant, the system can safely delete non-critical images while preserving diagnostic accuracy, thereby reducing checking time without compromising reliability.
2Reliability
If all images in a large image sequence are checked, then disease detection accuracy is maintained, but the time required increases significantly
Solution Approach 1:
The patent extracts representative images from groups of similar images and removes redundant images that provide no additional diagnostic information. By taking out only the essential images that contain unique information, the system maintains disease detection accuracy while significantly reducing the number of images users need to check.
Solution Approach 2:
The patent changes the parameter of image selection from individual image evaluation to group-based representative image selection. By calculating similarity metrics and identifying representative images that best represent each group, the system reduces the total number of images to be checked while preserving all diagnostically relevant information.
3Quantity of substance
If representative images are selected based on similarity, then the number of images is reduced, but the relationship between deleted and retained images may cause important areas to be missed
Solution Approach 1:
The patent performs preliminary calculation of similarity relationships between all images and identifies representative images before deletion. By pre-establishing the relationship map and determining which images are essential for complete area coverage, the system can reduce the number of images while ensuring no important areas are missed in the summarization process.
Data Source
AI summary
An image sequence acquisition section acquires an image sequence including a plurality of images. A processing section performs an image summarization process that acquires a summary image sequence based on first and second deletion determination processes that delete some of the images included in the acquired image sequence. The processing section sets an attention image sequence including one at least one attention image included in the plurality of images, selects a first reference image from the attention image sequence, selects a first determination target image from the plurality of images, and performs the first deletion determination process that determines whether the first determination target image can be deleted based on first deformation information that represents deformation between the first reference image and the first determination target image. The processing section sets a partial image sequence from the image sequence, a plurality of images that have been determined to be allowed to remain by the first deletion determination process being consecutively arranged in the partial image sequence. The processing section selects a second reference image and a second determination target image from the partial image sequence, and performs the second deletion determination process that determines whether the second determination target image can be deleted based on second deformation information that represents deformation between the second reference image and the second determination target image.