Image Fusion Device and Method Using Weighted Least Squares Filter and Sparse Representation

Medical Multimodal Image Fusion Device Based on Weighted Least Squares Filter and Sparse Representation

This technology is an image fusion device that decomposes different medical images, such as CT and MRI, into low-frequency and high-frequency layers using a Weighted Least Squares (WLS) filter. It then fuses the low-frequency layers using a Laplacian pyramid and sparse representation, and the high-frequency layers using a maximum absolute fusion rule.

Recently, it has become difficult to obtain sufficient diagnostic information from a single type of medical image, necessitating the combination of multiple images. However, existing fusion methods that directly manipulate pixels in the spatial domain are fast but suffer from spatial distortion and a lack of spectral information, leading to poor quality in the fused images.

This technology improves the quality of fused images near edges by calculating sparse coefficients of low-frequency patches using an orthogonal matching pursuit algorithm and a learned dictionary, selecting larger coefficients, and reconstructing them via an inverse Laplacian pyramid. It can be applied to multi-image-based clinical diagnosis and treatment planning systems, helping physicians interpret both anatomical and functional information from a single image.

‍

‍

Key Features:

• An image decomposition unit that uses a Weighted Least Squares filter to decompose input images into a low-frequency layer (smoothed image) and a high-frequency layer (difference image)
• A Laplacian pyramid decomposition unit that applies a Laplacian pyramid to the low-frequency layer to decompose it into two sub-low-frequency layers and a set of sub-high-frequency layers
• A sparse representation unit that rearranges sliding window patches into column vectors and calculates sparse coefficient vectors using an orthogonal matching pursuit algorithm and a dictionary learned via the least squares method
• A final image fusion unit that restores the final image by adding the fused low-frequency layer to the final high-frequency layer reconstructed using the maximum absolute fusion rule

‍