A ridgelet representation of semantic object using watershed segmentation

Abstract

We propose an application of the ridgelet transform to semantic objects which are produced by the watershed segmentation. The ridgelet transform is effective in representing line singularities; therefore, it is a powerful tool for coding. Moreover, it has the advantageous property of rotating a block easily by shifting coefficients in ridgelet domain. Nevertheless, the rotating target is not a rectangular block but a real object. For that reason, we divide a picture into semantic objects using the watershed segmentation, and each object is converted by the ridgelet transform; then, we can rotate each object easily. Predictive coding with this method is attempted. Simulations show that referencing object with rotation is very effective in prediction.