Publication Date: 2022/07/01
Abstract: Image reconstruction is currently often used in a wide range of technological and medical applications. The local image feature descriptor is the most critical factor influencing the performance of object reconstruction or image retrieval systems. This study provides and demonstrates a strategy for replicating images. In this approach, Training photos are used to extract local feature descriptors; at first images are recreated using local feature descriptors and geometric information. Scale invariant feature transform (SIFT) descriptors are used to characterize images, and the feature extraction method is similar to how descriptors are used in the training phase. The unknown image closest neighbor descriptor built by using pairwise matching. For each of the regions of interest, visually equivalent patches may be in the external image database.To detect patch overlapping regions between the new patch and the patch already present in the query image, the Mean Squared Error (MSE) is used. To eliminate overlapping patches, the highest MSE threshold value is chosen as the default threshold (DT) in this experimental technique. Based on the experimental results, an image may be approximated and rebuilt using image local feature descriptors.
Keywords: Image reconstruction, image retrieval, image feature descriptor, geometric information, partial information.
DOI: https://doi.org/10.5281/zenodo.6787485
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT22MAY2034_(3).pdf
REFERENCES