Multi-Scale Boundary Detection in Natural Images

In this work we empirically study the multi-scale boundary detection problem in natural images. We utilize local boundary cues including contrast, localization and relative contrast, and train a classifier to integrate them across scales. Our approach successfully combines strengths from both large-scale detection (robust but poor localization) and small-scale detection (detail-preserving but sensitive to clutter). We carry out quantitative evaluations on a variety of boundary and object datasets. We show that multiscale boundary detection offers significant improvements, ranging from 20% to 50%, over single-scale approaches. Our conceptually simple approach outperforms existing algorithms on the Berkeley Segmentation Benchmark.