Localised decorrelation stretch (L-DCS) for improved visibility of large, dense rock art scenes

Overview

This work addresses limitations in DStretch, a widely used image enhancement tool in rock art archaeology, by introducing Localised Decorrelation Stretch (L-DCS). DStretch applies global decorrelated stretching across entire images, which diminishes performance when RGB values exhibit wide distributions or when rock art assemblages contain densely painted areas with high local correlation. L-DCS partitions images into overlapping windows, applies decorrelated stretching locally within each window, and merges results to produce unified enhanced images. The method is designed to operate automatically on large datasets without requiring manual parameter adjustment per image.

Methods and approach

L-DCS implements a windowed approach to decorrelated colour stretching. The algorithm segments input images into spatially overlapping windows of specified dimensions, calculates decorrelated stretching parameters independently for each window based on local RGB statistics, applies the stretch transformation to pixels within that window, and blends results at window boundaries to eliminate discontinuities. The method requires specification of window size and overlap parameters. Testing compared L-DCS output against standard DStretch results across dense rock art scenes, orthomosaics, and three-dimensional model textures. Evaluation focused on retention of fine motif details and visibility of obscured or faint pictographic elements within densely painted archaeological contexts.

Results

L-DCS demonstrated superior performance relative to DStretch in multiple evaluation scenarios. Enhanced images retained greater fine-scale detail within individual motifs, particularly where conventional global stretching produced blurring or loss of definition. Visibility of previously obscured or faint motifs in densely painted areas improved substantially. The method accommodated greater variation in pigment coloration across single images without requiring application of multiple customised filters. Consistent results were achieved across heterogeneous image types and rock art assemblages, suggesting robust performance across diverse archaeological documentation contexts.

Implications

L-DCS addresses a significant technical limitation in the current standard for rock art image enhancement. Automated local decorrelation processing enables more efficient analysis of large archaeological image datasets, reducing time required for manual filter application and optimisation per image. The method's capacity to handle both dense pictographic scenes and varied pigmentation without parameter recalibration expands its applicability across different rock art traditions and preservation contexts. Implementation of local windowing approaches may generalise to other archaeological imaging enhancement tasks beyond rock art documentation.