Why JPEG Uses Zig-Zag Scanning for DCT Coefficients

In the JPEG image compression standard, an 8x8 block of pixels is converted into frequency components via the Discrete Cosine Transform (DCT) and then quantized. Rather than reading these 64 coefficients in standard row-by-row or column-by-column raster order, JPEG rearranges them using a diagonal zig-zag pattern. This article explains the technical mechanics behind this ordering, demonstrating how it groups spatial frequencies from lowest to highest to dramatically optimize subsequent Run-Length Encoding (RLE) and entropy coding steps.

Energy Compaction and Spatial Frequencies

The two-dimensional DCT transforms spatial pixel values into spatial frequency coefficients. The top-left corner of the resulting 8x8 matrix represents the lowest frequencies—starting with the Direct Current (DC) component at index (0,0), which holds the average brightness of the block. As you move away from the top-left toward the bottom-right corner, the coefficients represent progressively higher vertical, horizontal, and diagonal spatial frequencies (Alternating Current or AC components).

Natural images exhibit high spatial correlation, meaning most of the visual energy and perceptible information is concentrated in these lower frequencies.

The Effect of Quantization

Human vision is far less sensitive to subtle variations in high-frequency patterns than to broad luminance and low-frequency structures. Consequently, JPEG's quantization step intentionally uses larger divisors for high-frequency coefficients located toward the bottom-right. When the divided values are rounded to the nearest integer, the vast majority of medium- and high-frequency coefficients become zero. The non-zero values are predominantly clustered in the upper-left region.

The Zig-Zag Traversal

Reading the 8x8 grid in standard row-major order would alternate haphazardly between low-frequency non-zero values and high-frequency zeros. To align the coefficients in a predictable order of visual importance, JPEG reads them along diagonal paths:

1. It begins at the DC coefficient (0,0).
2. It proceeds diagonally back and forth across the matrix, traversing lower frequencies first.
3. It ends at the highest combined horizontal and vertical frequency at the bottom-right (7,7).

This path systematically sorts the 64 two-dimensional values into a one-dimensional array ordered almost entirely by monotonically increasing frequency.

Maximizing Compression with Run-Length Encoding (RLE)

Sorting by frequency directly serves JPEG's subsequent compression algorithms:

• Clustering Non-Zeros: Non-zero coefficients are pushed to the very beginning of the 1D stream.
• Long Sequences of Zeros: As the scan enters higher frequencies, the sequence quickly transitions into long runs of consecutive zeros.
• Run-Length Encoding Efficiency: JPEG encodes AC coefficients as pairs indicating (run of preceding zeros, value of next non-zero coefficient). Grouping zeros together minimizes the number of symbols that must be created.
• The End of Block (EOB) Marker: Once the last non-zero coefficient in the zig-zag scan has been read, JPEG emits a single, compact "End of Block" (EOB) code. This single marker signals the decoder that every remaining coefficient up to the 64th position is zero, allowing the encoder to skip storing dozens of individual zeros per block.

By restructuring the 2D frequency distribution into a predictable 1D gradient of non-zeros followed by zeros, the zig-zag scan drastically reduces the final byte size before Huffman or arithmetic coding takes place.