Statlog Image Segmentation Data for Classification

Statlog (Image Segmentation) is a classic dataset for benchmarking classification algorithms on image segmentation tasks. It contains 2,310 instances derived from outdoor images, where each instance represents a 3x3 pixel region described by 19 numerical attributes. The dataset is part of the UCI Machine Learning Repository, a known source for machine learning data.

Use Cases

• Classify the 'class' label (e.g., brickface, sky, foliage) using the 19 region-based feature attributes.
• Train a decision tree classifier to predict image segment categories from texture and edge-related features.
• Benchmark multi-class classification algorithms like SVM or Random Forest on the 7 distinct class labels.
• Perform feature importance analysis on the 19 attributes to identify key visual characteristics for segmentation.

Strengths

• Contains 2,310 instances for model training and evaluation.
• Provides 19 distinct numerical features per instance for analysis.
• Well-known benchmark with documented use in machine learning literature.

Limitations

• Dataset size of 2,310 instances may be limited for deep learning approaches.
• Features are pre-extracted tabular representations, not raw images.
• Temporal and geographic context for the original outdoor images is unspecified.

Provenance

Source

UCI Machine Learning Repository.

Collection Method

Features extracted from a database of 7 outdoor images, each manually segmented to create labeled 3x3 regions.