Normalized weighted cross correlation for multi-channel image registration

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Normalized Cross-Correlation (NCC) is a standard feature-agnostic method for image registration.
• Existing NCC methods have limitations with images of arbitrary dimensions, channels, and sampling irregularities.

Purpose of the Study:

• To generalize Normalized Cross-Correlation (NCC) by introducing pixel-wise weighting for enhanced image registration.
• To develop a generalized Normalized Weighted Cross-Correlation (NWCC) and its zero-mean variant (ZNWCC).
• To provide Discrete Fourier Transform (DFT) formulations for efficient computation of NWCC and ZNWCC.

Main Methods:

• Reviewed existing NCC definitions across various image dimensions and channels.
• Proposed a generalized NWCC allowing individual weighting of pixel values per channel.
• Developed DFT formulations for fast computation of NWCC and ZNWCC, including overlap calculation.

Main Results:

• Introduced generalized NWCC and ZNWCC enabling prioritization of pixels, e.g., by signal-to-noise ratio.
• Demonstrated that NWCC facilitates registration of uniformly sampled images with irregular boundaries and sparse sampling.
• Provided DFT-based computation methods for efficient implementation and overlap calculation.

Conclusions:

• The proposed generalized NWCC and ZNWCC offer a flexible and powerful approach to image registration.
• The DFT formulations ensure computational efficiency for practical applications.
• The enhanced methods improve the registration of challenging image datasets with non-uniform sampling.