ERIS Output Visualizations
ERIS Balanced Output
Processed using adaptive median thresholding to achieve statistical uniformity (appears as more uniform noise). This balanced data was used to pass tests like Dieharder/NIST STS that expect uniformity. Distinct from the cryptographically whitened `eris:full` output available via the API.
ERIS Pure Raw Output (`eris:none`)
Direct, unconditioned output (Conceptually `eris:none`). Contains inherent complexity and potential patterns/structures (visible as clustering or density variations). Despite visual non-uniformity, this raw output passes the Dieharder test suite (1GB data tested, 1 weak result noted), confirming its high degree of unpredictability.
ERIS Whitened Output (`eris:full`)
Output processed by the full whitening pipeline (e.g., XOR cascade + Toeplitz) designed to maximize statistical uniformity and cryptographic suitability. Appears as visually uniform noise.
Frequency Analysis (FFT)
A 2D Fast Fourier Transform (FFT) reveals the spatial frequency content of the visualizations. Bright areas in the FFT spectrum indicate dominant frequencies. Comparing raw and whitened data FFTs highlights the impact of the whitening process.
Raw Data (`eris:raw`) - Bit Visualization
Raw Data (`eris:raw`) - FFT Spectrum
The raw bit visualization exhibits strong diagonal banding, a common characteristic of unprocessed physical entropy sources. The FFT spectrum clearly shows this structure as prominent vertical bands, indicating significant spatial correlation at specific frequencies. This visual non-uniformity is expected before whitening.
Whitened Data (`eris:full`) - Bit Visualization
Whitened Data (`eris:full`) - FFT Spectrum
After processing through the full whitening pipeline, the strong diagonal patterns are removed, resulting in a visually more uniform bit field. However, the FFT spectrum reveals that faint, large-scale (low-frequency) variations remain, visible as a concentration of energy near the center. This demonstrates how whitening optimizes for statistical tests but can leave subtle spatial frequency artifacts.
Wavelet Analysis (Multiscale)
A 2D Discrete Wavelet Transform (DWT) decomposes the image into different frequency components localized in space and scale. The top-left square (LL) represents the coarsest approximation, while surrounding squares show detail coefficients (horizontal, vertical, diagonal) at progressively finer scales. This helps analyze features across scales.
Raw Data (`eris:raw`) - Wavelet Decomposition
The wavelet decomposition of the raw data clearly shows structure in the detail coefficients (the gray areas outside the top-left approximation square). Note the distinct diagonal banding textures, particularly noticeable in the horizontal (LH, top-right sections) and diagonal (HH, bottom-right sections) detail bands across different scales. This intricate, directional structure corresponds to the patterns seen in the raw bit visualization and is expected before whitening.
Whitened Data (`eris:full`) - Wavelet Decomposition
After whitening, the detail coefficients (surrounding gray areas) appear significantly more uniform and noise-like, confirming the suppression of the raw data's structures. However, close inspection reveals very faint, low-contrast residual textures within these detail bands. The final approximation (top-left square) is smooth, indicating the large-scale artifacts seen in the FFT are not strong localized features at the coarsest wavelet scale.