선택한 단어 수는 입니다.
최소 단어 이상 선택하여야 합니다.
최소 단어 이상 선택하여야 합니다.
선택한 단어 수는 30입니다.
최대 10 단어까지만 선택 가능합니다.
최대 10 단어까지만 선택 가능합니다.
제안 방법
- Since a pair of the extracted matching points doesn’t always locate in the most outer point, we define α as an error range in Eq. (1) The existing method may calculate the matching point for all the pixels, while the proposed method calculates the partial scan by automatically deciding the scan area.
- If the quadrant without a pair of SIFT matching points exists, then we examine whether the edge in the quadrant exists. If the edge exists, then the proposed method classifies the region into foreground area and compensates the disparity with the average disparity of the SIFT matching points existing in each of the quadrant. ‘S’ in Fig.
- In the experiment, we used stereo images of a building model with 320 × 240 sizes and windows with 21 × 21 sizes to measure the area-based similarity.
- In this paper, we propose an efficient method to generate a correct disparity map for a 3-dimensional reconstruction. Generally, algorithms[1] for generating disparity maps can be classified into two categories: region-based method[2] and feature-based method[3].
- Existing stereo matching still has unsolved problems for searching accurate corresponding points and compensating occlusion region generated by the absence of the corresponding points. In this paper, we propose the method to solve problems for occlusion region by using SIFT matching points.
- However, the accuracy measure is difficult because the reference disparity map isn't defined. In this paper, we use the SIFT matching pairs to measure the accuracy of the proposed method. Because the SIFT algorithm which is robust to the changes of the image scaling and rotating has high trust about the matching pair, we use the disparity extracted by the SIFT matching pair as a reference disparity map.
- That is, the proposed method automatically classifies an image into background and object area for matching calculation. This method has an advantage that can reduce the calculation quantity due performing to the matching calculation only for object area except background area.
- SIFT-based matching algorithm is strong for rotating and resizing the image, and can improve reliability for searching a pair of the matching points. That is, this algorithm select a pair of the matching points with the closest distance among SIFT matching points existing in the quadrant of the 2D coordinate system which is made from a pixel point of occlusion region as the origin. If the quadrant without a pair of SIFT matching points exists, then we examine whether the edge in the quadrant exists.
- 8-(d) indicates the result after compensating occlusion region by eliminating errors for false matching. The proposed method compensates the occlusion region using the similarity of adjacent pixels and the accuracy of SIFT matching pairs. Fig.
- As a matching method, we use the proposed MMAD algorithm which is a modification of the existing MAD algorithm and the SIFT feature points robust to changes of image scale and rotation. The proposed method reduces errors about depth information through eliminating false matching errors by calculating the vector with SIFT and compensates the occluded regions by using pair of adjacent SIFT matching points.
- Generally, algorithms[1] for generating disparity maps can be classified into two categories: region-based method[2] and feature-based method[3]. The proposed method uses a region based similarity measurement as a basis matching technique and improves correctness using characteristic method based on the extracted matching points from SIFT[4]. In order to automatically establish scan area for matching between two disparity maps, we extract a pair of matching points with the minimum and maximum value from SIFT.
이론/모형
- In this paper, we propose the algorithm considering both methods. In order to improve the matching accuracy, the proposed matching method uses region-based method for removing errors and utilizes feature-based method for compensating false area.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.