Understanding how different image formats store data internally reveals why certain formats are better suited for specific use cases. Each format represents a different engineering tradeoff between file size, quality, feature support, and compatibility.

PNG (Portable Network Graphics) uses lossless compression based on the DEFLATE algorithm, the same compression used in ZIP files. Internally, a PNG file is organized into chunks, each with a four-character type code. The critical chunks include IHDR (image header with dimensions and color type), IDAT (the compressed image data), and IEND (end marker). Optional chunks like tEXt store metadata, tRNS handles transparency for palette-based images, and iCCP embeds color profiles. Before compression, PNG applies a prediction filter to each row of pixels (None, Sub, Up, Average, or Paeth) that transforms pixel values into differences from predicted values, which compress more efficiently. This is why PNG excels at images with large areas of uniform color but produces large files for photographs.

JPEG compression is fundamentally different, using a lossy approach based on the Discrete Cosine Transform (DCT). The image is first converted from RGB to YCbCr color space, separating luminance (brightness) from chrominance (color). The chrominance channels are typically downsampled to half resolution because human vision is less sensitive to color detail than brightness detail. Each channel is then divided into 8x8 pixel blocks, and the DCT transforms each block from spatial pixel values into frequency coefficients. A quantization step divides these coefficients by values from a quantization table and rounds to integers, permanently discarding high-frequency detail that is less perceptible to the human eye. This quantization step is where quality loss occurs, and the quality slider controls how aggressively coefficients are quantized.

WebP, developed by Google, uses two distinct approaches. Lossy WebP is based on VP8 video codec technology, using predictive coding where each block is predicted from already-decoded blocks and only the prediction error is encoded. Lossless WebP uses a completely different algorithm with techniques including spatial prediction, color space transform, backward reference using LZ77, and entropy coding. WebP typically achieves 25-35% better compression than JPEG at equivalent visual quality. The mathematics behind lossy versus lossless compression relates to information theory: lossless compression can only exploit statistical redundancy in the data (achieving maybe 2:1 to 3:1 compression for photos), while lossy compression additionally exploits perceptual redundancy by removing information the human visual system cannot easily detect, achieving 10:1 or higher compression ratios.