A lossy compression technique that rounds similar values to a single representative value. For example, pixel brightness values of 119, 120, 121, and 122 might all become 120. The data becomes more repetitive and compresses better, but the original values can't be recovered.