An image stitching algorithm with improved Sigmoid function weights
Haojie Cao, Xu Zhang · 2023
Image stitching can combine multiple adjacent images with overlapping areas into a single high-resolution panoramic image that depicts the complete information about the target scene, this technique can be well applied to the visual measurements of optical profilometry instrument. However, there may still be relatively obvious splicing traces at the edges of the overlapping areas after image stitching with traditional weighted average fusion algorithm or linear weighted gradual fading-in and fading-out fusion algorithm. To address this issue, this paper proposes an image stitching algorithm that uses the improved Sigmoid function to calculate the non-linear weights. Firstly, the extended phase correlation operation and image resampling are used for image registration, and then the weight information of the target image is calculated by using the improved Sigmoid function, in other words the original Sigmod function is scaled, so that the function is shifted right and stretched as a whole, and the domain is set as the left boundary of the overlapping area to the right boundary of the overlapping area, the range corresponding to the domain is regarded as the weight information of the target image, the weight of the reference image and the weight of the target image satisfy the relationship of summing to one. Finally, the calculated two non-linear weights are used for image stitching. The experimental results show that the improved image fusion algorithm works better whether the camera is shooting with or without exposure differences.