Forensic Dashboard
No Image Loaded
1. Log Visuals vs. Linear Truth: The images use Log scaling so you can see them. Internally, we score using Linear power to catch "invisible" grid spikes that are mathematically massive.
2. Why Laplacian? We run a High-Pass filter to strip image content. We are looking for the manufacturing defects of the Generator, not the picture itself.
3. Artifact Identification:
- Star Field / Grid: Indicates upsampling artifacts (Checkerboard effect) common in GANs and older Diffusion.
- Solid Halo: Indicates heavy post-processing or compression.
- Natural Falloff: Real photos usually look like a chaotic cloud with no distinct geometric patterns.
Technical Methodology & Signal Processing
1. The Pipeline
Unlike simple metadata checkers, this tool performs frequency domain analysis. The image is processed in four stages:
Input Image -> Blue Channel Extraction -> Laplacian High-Pass -> Windowing -> FFT
2. Channel Selection
We default to the Blue Channel. In digital sensors (Bayer Filter), the Blue channel is the least dense (25% of pixels) and usually has the lowest signal-to-noise ratio. It is the "trash can" of the image where compression artifacts and generation errors are most prominent.
3. Edge Detection (Laplacian)
A Laplacian kernel (3x3) acts as a second-order derivative filter. It removes the "DC Component" (the average brightness and color) and leaves only the rate of change. This effectively removes the "picture" (the face, the tree) and leaves the "texture" (the pixel relationships).
4. Frequency Domain (FFT)
We use a Fast Fourier Transform to convert the spatial data (pixels) into frequency data (sine waves).
The Center: Low frequencies (gradual gradients).
The Edges: High frequencies (sharp noise/edges).
Generative AI, specifically Convolutional Neural Networks (CNNs), often utilize "Transposed Convolutions" to upscale images. This creates a "Checkerboard Artifact" which, while invisible to the naked eye, manifests as a distinct periodic grid in the frequency domain.