JPEG Crop Impact Study

Study how cropping (reducing the image area while keeping the analysis fragment at full resolution) affects JPEG compression quality. Unlike the resolution-impact study which downscales the entire image, this study crops the image to different sizes around a fixed 200×200 analysis fragment, eliminating the confounding effect of reduced resolution on quality metrics. Quality metrics are measured only on the analysis fragment, ensuring consistent comparison across crop levels.

Dataset: div2k-valid (100 images) · 3465 measurements · Formats: JPEG

Tool Versions
ToolVersion
ssimulacra2available
butteraugliavailable
ffmpeg7.1.3-0+deb13u1
cjpeg2.1.5
cwebp1.5.0
avifenc1.3.0
cjxl0.11.2

SSIMULACRA2 vs BPP

↑ Back to top

BUTTERAUGLI vs BPP

↑ Back to top

PSNR vs BPP

↑ Back to top

SSIM vs BPP

↑ Back to top

SSIMULACRA2 vs Crop

↑ Back to top

BUTTERAUGLI vs Crop

↑ Back to top

PSNR vs Crop

↑ Back to top

SSIM vs Crop

↑ Back to top

BITS PER SSIMULACRA2 PER PIXEL vs Crop

↑ Back to top

BITS PER BUTTERAUGLI PER PIXEL vs Crop

↑ Back to top

BITS PER PSNR PER PIXEL vs Crop

↑ Back to top

BITS PER SSIM PER PIXEL vs Crop

↑ Back to top

BPP vs Crop

↑ Back to top

ENCODING TIME PER PIXEL vs Crop

↑ Back to top

Visual Comparison

Side-by-side comparison of encoding artifacts on the most representative image fragments, selected automatically from the dataset using Butteraugli spatial distortion analysis.

Anisotropic Variance Analysis

Identifies the image and fragment where the choice of encoding parameters has the greatest anisotropic impact on quality. One fragment is selected per target metric group and shared across all target values within that group for direct comparability.