Image Fusion Metrics

Image fusion is a crucial technology in computer vision and image processing, aiming to integrate information from multiple source images to generate a higher-quality composite image. To assess fusion performance, researchers have developed various quantitative metrics that measure the quality of fused images from different perspectives.

Joint Entropy Joint entropy measures the information content in a fused image. Higher entropy values indicate richer information content, usually suggesting better fusion results. Implementation involves analyzing the grayscale histogram of the image to assess information quantity. Code implementation typically uses histogram calculation functions like numpy.histogram() followed by entropy computation.

Spatial Frequency Spatial frequency reflects the clarity and detail level of an image, including row frequency and column frequency. More high-frequency components indicate better preservation of edge and texture information. Algorithm implementation calculates gradients in both horizontal and vertical directions using convolution operations.

Mutual Information (MI) Mutual information measures the degree of information sharing between the fused image and source images. Higher MI values indicate that the fused image better preserves critical information from the source images. Implementation involves joint probability distribution calculation between images using histogram-based methods.

Structural Similarity (SSIM) Evaluates the similarity between fused and source images in terms of structure, luminance, and contrast. Values closer to 1 indicate better structural consistency. The algorithm computes mean, variance, and covariance of local image windows using sliding window approach.

Peak Signal-to-Noise Ratio (PSNR) Measures the error between fused and reference images. Higher PSNR values indicate less distortion. Calculation uses mean squared error (MSE) followed by logarithmic scaling relative to maximum pixel value.

Edge Preservation Detects whether the fused image adequately preserves edge information from source images, typically computed using edge detection operators like Sobel or Canny. Implementation involves gradient magnitude calculation and thresholding operations.

Visual Fidelity Incorporates human visual system characteristics to assess whether the fused image appears natural visually. This metric often uses perceptual models that simulate human vision mechanisms.

These metrics comprehensively evaluate fusion algorithm performance and are applicable to multimodal image fusion (such as infrared-visible light fusion) or super-resolution reconstruction tasks. By comparing metric scores across different algorithms, researchers can objectively select optimal fusion strategies. Implementation typically involves Python libraries like OpenCV, scikit-image, or MATLAB's Image Processing Toolbox.