Enhancing registration precision of multispectral breast images by fusing multi-wavelength information based on an improved gradient descent method

Multispectral transmission imaging is widely recognized as a developmental approach for early breast cancer screening. However, a persistent challenge exists in image registration, particularly in processing human images, where registration often relies solely on single-wavelength information, leading to inadequate accuracy. This paper purposes a novel image registration method that involves summing pixel values from three color channels at identical pixel coordinates and utilizes an improved gradient descent algorithm to achieve image registration. Multispectral maps at wavelengths of 445 nm, 520 nm, and 620 nm serve as examples for testing and evaluation, with frames summed for registered images at each wavelength. Experimental results demonstrate the superior registration accuracy of this method compared to registering images of a single-wavelength independently. This advancement enhances the accuracy and reliability of multispectral transmission imaging in the early detection of breast cancer.