Multi-Focus Image Fusion Based on Three-Channel Inseparable Symmetric Wavelet and Its Relevance to NDT Image Analysis

Literature Overview

The paper by Liu Bin, Liu Weijie, and Ma Jiali, published in the Chinese Journal of Scientific Instrument (Volume 33, Issue 5, 2012, pp. 1110–1116), presents a novel multi-focus image fusion method based on a three-channel inseparable symmetric wavelet transform. While this work originates in the field of signal processing and image analysis, it carries significant relevance to non-destructive testing (NDT) applications in the steel pipe and welding industry, where the fusion of multiple NDT images is a growing practical need. The research was supported by the National Natural Science Foundation of China (Grant 61072126) and the Hubei Provincial Natural Science Key Fund (Grant 2009CDA133).

Core Technical Methodology

The Problem with Conventional Wavelet-Based Fusion

The authors identify two fundamental limitations of conventional Daubechies wavelet-based image fusion methods:

1. Lack of symmetry: Daubechies wavelets are inherently asymmetric, which introduces phase distortion in the reconstructed fused image.
2. Directional limitation of tensor product transforms: Traditional tensor product wavelet transforms only emphasize horizontal and vertical directions, missing diagonal features that are critical in NDT images (e.g., crack orientations, weld seam boundaries).

Three-Channel Inseparable Symmetric Wavelet Construction

The key innovation is the construction of a three-channel inseparable symmetric wavelet filter bank using a matrix expansion method. The "inseparable" property means that the transform is not a simple tensor product of 1D filters, allowing it to capture directional features in multiple orientations. The "symmetric" property ensures that the filter bank preserves phase information, leading to better spatial alignment in the fused output.

Fusion Rules

The proposed fusion method employs a non-downsampling multi-scale decomposition, followed by coefficient-level fusion using the following rules:

• Low-frequency coefficients: The minimum value is selected (to preserve background and reduce noise)
• High-frequency coefficients: The maximum absolute value is selected (to preserve edge and defect information)

This combination of minimum-value selection for low frequencies and maximum-absolute-value selection for high frequencies ensures that the fused image retains both the smooth background from all input images and the sharpest edges and most prominent features.

Relevance to NDT and Welding Quality Inspection

Application to Multi-Method NDT Image Fusion

In modern welding quality inspection, it is common to employ multiple NDT methods on the same weld or fitting, each producing images with complementary information:

The multi-focus image fusion technique described in this paper could be applied to fuse images from different NDT methods, creating a comprehensive defect map that combines the strengths of each method. For example, fusing an RT image (good for volumetric defects) with a PAUT image (good for planar defects) could provide a more complete picture of the defect population in a weld.

Application to Multi-View Inspection Imaging

In the inspection of pipe fittings such as tees, elbows, and reducers, the complex geometry often requires multiple inspection views. The fusion of images from different viewing angles could provide a more complete representation of surface and near-surface defects. The symmetric wavelet approach ensures that features at the boundaries of the inspection field are not distorted, which is critical for accurate defect characterization.

Application to Multi-Focus Documentation Imaging

In engineering documentation and as-built surveys, multi-focus images of pipe racks, weld joints, and fitting installations are often needed. The fusion technique described here could be applied to create single, high-quality documentation images that combine the sharp focus of different regions in the scene.

Study Insights and Reflections

As a metallurgist and welding engineer, I appreciate the potential of this image fusion technique for enhancing NDT capabilities. The quality of NDT is fundamentally limited by the information content of the inspection images, and better image fusion methods directly translate to better defect detection and characterization. The symmetric wavelet approach is particularly attractive because phase distortion in NDT images can lead to misinterpretation of defect location and size, which are critical parameters for fitness-for-service assessment.

However, I would note that the practical implementation of this technique in industrial NDT settings faces several challenges. First, the computational requirements of the three-channel inseparable wavelet transform may be significant, and real-time processing may be difficult to achieve on portable inspection equipment. Second, the fusion rules described (minimum for low frequencies, maximum absolute value for high frequencies) are relatively simple and may not be optimal for all NDT image types. For example, in radiographic images, the defect signal is typically a reduction in film density (a dark region), and the fusion rule would need to be adapted accordingly.

The study represents a valuable contribution to the field of image processing, and I believe that further collaboration between signal processing researchers and NDT practitioners could lead to practical implementations that improve the reliability of welding quality inspection. The key to such collaboration would be the development of NDT-specific fusion algorithms that account for the physics of each inspection method and the statistical properties of defect signals.