Mask-based image blending and its applications on mobile devices
Yingen Xiong, Kari Pulli · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2009
A mask-based image blending approach is presented and implemented on mobile devices for producing mobile panoramic images. In this approach, a single channel mask is created and initialized with distributed values for each source image. It is warped and interpolated together with its source image during panorama stitching to keep related transformation information. The values of the mask provide weighting coefficients for blending images together to produce a panoramic image. Compared to other complicated approaches such as gradient domain image blending, its computational and memory cost is low. Compared to other simple approaches such as alpha blending, it does not need to locate boundaries of overlapping areas for determining weighting coefficient values. It can also be applied to 2D panorama stitching and its blending quality is better. The approach is implemented in a mobile panorama system to produce panoramic images for preview. It shows good performance in processing image sequences captured in both indoor and outdoor scenes. The property in low computational and memory costs as well as fast blending speed of the approach really benefits the mobile panorama system.