研究目的
To propose an efficient image denoising algorithm that combines block-based operations, rotation-based SVD, and wavelet transform to improve denoising performance by considering directional features and local stationarity of images.
研究成果
The proposed block-rotation-based SVD filtering in the wavelet domain is effective for image denoising, showing superior performance in PSNR and SSIM compared to conventional methods like wavelet thresholding and standard SVD. It is particularly beneficial for images with strong noise and distinct textures, but its application is limited to specific image types and requires significant computation.
研究不足
The method has higher computational costs compared to traditional SVD methods and is primarily suitable for square gray images with distinct directional textures.
1：Experimental Design and Method Selection:
The method involves dividing the noisy image into non-overlapping sub-blocks, applying a single-level discrete 2-D wavelet transform to each sub-block to separate low-frequency and high-frequency parts, using SVD with rank-1 approximation for noise filtering in horizontal and vertical high-frequency parts, and employing rotation-based SVD for diagonal high-frequency parts by rotating 45 degrees before filtering and back after. The inverse wavelet transform is used for reconstruction.
2：Sample Selection and Data Sources:
Test images of size 512x512 with added white Gaussian noise (mean 0, normalized variance from
3：001 to 1) are used. List of Experimental Equipment and Materials:
No specific equipment or materials are mentioned; the method is implemented computationally, likely using software like MATLAB.
4：Experimental Procedures and Operational Workflow:
Steps include image division, wavelet decomposition, SVD filtering for horizontal/vertical parts, rotation-based SVD for diagonal parts, inverse wavelet transform, and image reorganization.
5：Data Analysis Methods:
Performance is evaluated using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
