"The Glitching Hour" by Amy Wibowo

### Overview In this talk, Amy Wibowo (known online as sailor HG) explores digital aesthetics, focusing on how image compression mechanisms can be understood and manipulated to create artistic "glitch" effects. Paralleling how digital filters recreate analog camera imperfections like light leaks and film grain, she examines how digital compression artifacts and file corruption can produce nostalgic visual aesthetics.

### Key Topics - **Digital Image Fundamentals and Bitmap Glitching**: A basic digital image consists of color samples (pixels) arranged in rows and columns, typically represented as red, green, and blue (RGB) values or hue, saturation, and value. Straightforward uncompressed formats like BMP precede sequential pixel data with headers (width, height, color space). In raw pixel data, glitches can be achieved directly by sorting pixel rows/columns, shifting spatial sections, or shifting individual color channels. - **Lossless vs. Lossy Compression**: Compression reduces storage space while preserving discernible meaning. Lossless compression removes redundancies (such as counting consecutive identical characters) and allows exact reconstruction. Lossy compression discards details that are hard to notice (analogous to stripping vowels and punctuation from text). - **JPEG Color Channel Separation**: JPEG reduces file size by dividing image details into varying layers. Splitting RGB into luminance, blue-green, and red-green channels takes advantage of human visual perception, which is much less sensitive to color shifts than brightness. Because luminance contains the primary detail, color channels can be stored at half resolution with minimal perceptible degradation. - **Discrete Cosine Transform (DCT) & Quantization**: Using an analogy of separating a string quartet recording into frequency tracks (violins, viola, cello) via Fourier analysis, the speaker explains how images behave like 2D waveforms. A Discrete Cosine Transform decomposes 8x8 pixel blocks into 64 2D basis functions. High-frequency coefficients (representing fine details) typically contribute less and are zeroed out or reduced by dividing by a quantization matrix determined by the JPEG quality setting.

### Technical Details - **JPEG Artifacting**: Setting the JPEG quality setting very low discards high-frequency detail aggressively. Because each 8x8 block is transformed independently, block boundaries misalign and become visibly prominent, while sharp edges approximated by cosine sums exhibit halos or oscillations. - **Targeted Byte Corruption**: Glitching compressed JPEGs without rendering the file unreadable requires avoiding image headers and metadata. Modifying only the compressed image stream—found after the `FF DA` section marker—ensures file decoders still parse the image. - **Renderer Differences and Error Propagation**: Replacing a random byte in the compressed payload produces vastly different outcomes depending on how specific image decoders handle invalid data: - In Safari, corruption errors often terminate after a single row. - In Chrome, errors propagate across the rest of the image because the baseline (0,0) DC component of an 8x8 block is encoded differentially from previous blocks. - Byte corruption can alter location values, displacing blocks, or hit separate color channels to cause sweeping hue shifts. - Decoders like macOS preview render exposed basis function patterns differently.

### Takeaways Understanding the mathematical foundations of JPEG compression—from color space subsampling and 8x8 DCT quantization to differential block encoding and file markers (`FF DA`)—allows creators to intentionally induce precise aesthetic glitches. Blending intentional quantization artifacting with targeted compressed-byte corruption enables complex artistic effects across web and desktop decoders.