Image Seam Estimation Method Using Downscaling and Cost Functions

Fast Seam Finding for Panorama Stitching Using Downsampling-Based Cost Functions

This technology identifies an approximate seam by downsampling two images to be stitched, then estimates the final seam by calculating pixel gradient-based costs and directional information within a limited area around the seam in the restored, full-resolution images.

When creating panoramic videos from sources like CCTV or action cameras, it is necessary to remove seams and ghosting artifacts after image registration. However, searching for an optimal seam across the entire high-resolution overlap area creates significant computational load and processing time, hindering real-time performance.

This technology reduces the search range at a low-resolution stage, calculates cumulative costs only within a specific pixel range around the interpolated seam, and determines the final seam by backtracking directional information from the minimum cost position in the bottom row. It can be applied to real-time stitching software for multi-camera surveillance and 360-degree imaging, enabling the fast generation of seamless panoramas even on low-spec hardware.

‍

‍

Key Features:

• Detecting a seam from downsampled versions of the two images to be stitched
• Setting a limited area around the interpolated seam within the overlap region of the upsampled images
• Estimating the final seam using costs calculated from corresponding pixel data and gradient values, along with seam direction information
• Calculating the cumulative cost for each row and tracing back the direction information from the minimum cumulative cost position in the bottom row to the top row

‍