Following the instructions for the "basic" pb-lite, the results below were obtained:
The F-score obtained was 0.60. The resultant contours are below:
I also explored the use of the Earth Mover's Distance on pblite. For the grayscale image, my cost metric was the distance between successive gray values. (eg, Between the bin for gray==40 and gray==48, the cost was 8.) For the texons, my cost metric was the Euclidean distance between the centroid texon values as computed by k-means for each bin.
The F-score obtained was 0.60. Very little difference qualitatively or quantitatively was seen in the results. This implies that the cost metrics used for the earth-mover's distance were, in practice, close enough to constant (as in the chi-squared distance) that they had little effect on the results. (The output files, however, do differ.) The resultant contours are below:
| Original pb-lite | pb-lite with earth-mover's distance | Sobel baseline | Canny baseline |