NÂNG CAO CHẤT LƯỢNG ẢNH BẰNG KẾT HỢP THUẬT TOÁN TRUYỀN THỐNG LUCY – RICHARDSON – ROSEN VÀ KỸ THUẬT HỌC SÂU DẠNG UNET=IMAGE QUALITY IMPROVEMENT BY COMBINING TRADITIONAL LUCY – RICHARDSON – ROSEN ALGORITHM AND UNET-FORMED DEEP LEARNING TECHNIQUE

0 Người đánh giá. Xếp hạng trung bình 0

Tác giả: Văn Sang Đinh, Trọng Hiếu Lê, Văn Nhu Lê, Quang Thi Nguyễn, Trọng Thắng Trần

Ngôn ngữ: vie

Ký hiệu phân loại:

Thông tin xuất bản: Tạp chí Khoa học và Công nghệ - Đại học Thái Nguyên, 2024

Mô tả vật lý: tr.143 - 150

Bộ sưu tập: Metadata

ID: 485610

Thêm vào giỏ

Các thuật toán khôi phục truyền thống đã cải thiện đáng kể chất lượng ảnh về độ phân giải, tính tương tự với ảnh gốc và độ tương phản. Trong đó, thuật toán truyền thống Lucy – Richardson – Rosen là một trong những phương pháp mới được phát triển gần đây cho cải thiện độ phân giải ảnh và tính tương đồng với ảnh gốc. Tuy nhiên, các nghiên cứu cũng chỉ ra rằng phương pháp này vẫn đưa đến tạp chất gây nhiễu trên ảnh khôi phục, xuất hiện các artifacts bám theo các chi tiết ảnh. Để loại bỏ được nhiễu này và nâng cao chất lượng ảnh khôi phục, bài báo này đã đề xuất sử dụng thêm kỹ thuật học sâu. Mô hình học sâu được sử dụng dạng Unet. Quá trình mô phỏng trên bộ ảnh y tế đã chỉ ra rằng với một ảnh y tế mờ từ hệ thống quang học, sau hai bước xử lý bằng thuật toán Lucy – Richardson – Rosen và kết hợp với mạng học sâu Unet đã cho ảnh khôi phục tốt hơn. Ảnh khôi phục đã giảm được thành phần nhiễu artifact, cải thiện độ phân giải và tăng tính đồng dạng với ảnh gốc được minh chứng bằng chỉ số tương đồng cấu trúc và chỉ số nhận thức thị giác của người.The contrast, similarity to the original image, and visual quality have been significantly improved with traditional restoration techniques. Among these, the conventional Lucy - Richardson - Rosen - Algorithm is one of the recently created algorithms for boosting resolution and image resemblance to the original. However, tests have found that this method still inserts noise into the recovered image, with artifacts following the image details. This research suggests using additional deep learning approaches to reduce noise and improve the quality of the recovered image. Unet is the deep learning model that was employed. The simulation process results on a set of medical images showed that with a blurred medical image from an optical system, after two processing steps using the traditional Lucy - Richardson - Rosen - Algorithm and combining it with the Unet deep learning network, the reconstructed image was better. The Structural Simililarity Index and Learned Perceptual Image Patch Similarity Index demonstrated that the reconstructed image had lower artifact noise component, better resolution, and a higher degree of resemblace with the original image.

1. Đọc trực tuyến

1. Đọc trực tuyến

Tạo bộ sưu tập với mã QR