CT Image Object Extraction via Sequential Segmentation and Carving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern Explosive Detection Systems (EDS) face challenges in accurately extracting objects from CT images due to limitations in Automatic Threat Recognition (ATR) algorithms, which struggle with segmentation and identification of objects with varying densities and sizes in checked baggage, leading to over- or under-segmentation and erroneous density calculations.
Innovation Solution
The proposed solution involves a three-stage algorithm for object extraction from CT images, including preprocessing to mitigate artifacts, sequential segmentation and carving (SC) using symmetric region growing and bilateral filtering, and post-processing to split and merge objects based on homogeneity and spatial proximity, ensuring accurate segmentation of objects with mean modified Hounsfield unit values greater than or equal to 500 and volumes greater than 50 cc.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sequential segmentation and carving is applied to extract objects from CT images, then segmentation accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the object extraction process into multiple sequential stages, each targeting specific object types with distinct density characteristics. The first stage extracts high-density objects (metallic items) using thresholding, the second stage extracts medium-density objects (organic materials) using region growing, and the third stage extracts low-density objects (inorganic materials) using carving. This multi-stage segmentation approach improves segmentation accuracy by treating different object types with specialized algorithms rather than using a single complex algorithm for all objects.
2Measurement precision
If multiple sequential segmentation stages are used to distinguish objects with varying densities, then object identification accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies preliminary action by performing preprocessing operations before the main segmentation process. Bilateral filtering is applied to reduce noise and enhance edges in the CT images before segmentation begins. Artifact reduction techniques are also applied in advance to mitigate beam hardening and streak artifacts. These preliminary actions improve the quality of input data for subsequent segmentation stages, thereby improving object identification accuracy while reducing the computational burden during the actual segmentation process.
Solution Approach 2:
The patent applies periodic action through its three-stage segmentation process, where each stage operates sequentially on different density ranges. The first stage processes high-density objects, the second stage processes medium-density objects, and the third stage processes low-density objects. This periodic, staged approach allows the system to focus computational resources on extracting specific object types at each stage rather than processing all objects simultaneously, improving both accuracy and efficiency.
3Manufacturing precision
If symmetric region growing and bilateral filtering are applied, then segmentation precision is improved, but computational resources required increase
Solution Approach 1:
The patent applies local quality by using bilateral filtering, which preserves edges while smoothing regions. The filter selectively processes pixels based on their local characteristics, applying stronger smoothing in homogeneous regions and preserving boundaries where intensity changes occur. This localized processing approach improves segmentation precision by maintaining edge integrity while reducing noise, and it optimizes computational resources by avoiding uniform heavy processing across the entire image.
Data Source
AI summary
Extracting objects from a computed tomography (CT) image, including: sequentially applying segmentation and carving on volumetric data of the objects in the CT image; and splitting and merging the segmented objects based on homogeneity of the objects in the CT image.