Crawling Noise in JPEG Compressed Video Explained

Crawling noise in flat background areas of JPEG-compressed video occurs when minute sensor noise causes low-frequency DCT coefficients to alternate unpredictably across quantization thresholds between consecutive frames. Because JPEG and Motion JPEG process each frame independently without temporal compression, flat surfaces lack high-frequency texture to mask these micro-fluctuations. This causes adjacent 8x8 pixel blocks to randomly shift in tone from one frame to the next, creating a visible churning or crawling illusion.

Discrete Cosine Transform and Block Division

JPEG compression operates by splitting an image into 8x8 pixel blocks and converting their spatial pixel data into frequency components using the Discrete Cosine Transform (DCT). In a completely flat or smooth area, such as an out-of-focus wall or a clear sky, almost all the visual information is concentrated in the DC coefficient, which represents the average brightness and color of the entire 8x8 block. Very few low-frequency alternating current (AC) coefficients are retained, and high frequencies are discarded entirely.

Sensor Noise and Quantization Thresholds

Camera sensors produce a baseline level of electronic and photon shot noise, meaning the exact pixel values in a static flat background vary slightly from frame to frame. While these variations are often imperceptible in uncompressed footage, JPEG compression relies on a quantization table to round DCT coefficients into discrete integer steps to save space.

When a background value hovers near a quantization boundary, sensor noise repeatedly pushes the DCT coefficient back and forth across that threshold:

• Frame 1: The coefficient rounds down to a lower integer value, rendering the block slightly darker.
• Frame 2: A tiny fraction of sensor noise raises the average value, rounding the coefficient up to the next integer, rendering the block slightly brighter.

Because this rounding happens independently across thousands of adjacent 8x8 blocks, the borders and brightness levels of these blocks flicker intermittently over time.

The Absence of Temporal Compression

Standard video codecs like H.264 or HEVC use inter-frame (temporal) prediction to compare adjacent frames and smooth out static backgrounds. JPEG and Motion JPEG (M-JPEG) are purely intra-frame formats; each frame is compressed in total isolation with zero awareness of previous or subsequent frames. Consequently, the encoder does nothing to enforce temporal consistency, allowing independent quantization decisions to persist unrestricted from frame to frame.

Masking Effects and Human Visual Perception

The human visual system is exceptionally sensitive to motion and contrast changes in smooth, uniform regions. In complex, highly textured areas—such as foliage or fabric—the high-frequency detail provides "visual masking," hiding quantization jumps beneath natural surface complexity. In a flat background, there is no spatial texture to distract the eye. As adjacent 8x8 blocks continuously snap between different quantized brightness states, the viewer perceives this static, localized block flickering as an organic, crawling texture moving across the screen.