Region-based Image Segmentation by Watershed Partition and DCT Energy Compaction

Abstract

An image segmentation approach by improved watershed partition and DCT energy compaction has been proposed in this paper. The proposed energy compaction, which expresses the local texture of an image area, is derived by exploiting the discrete cosine transform. The algorithm is a hybrid segmentation technique which is composed of three stages. First, the watershed transform is utilized by preprocessing techniques: edge detection and marker in order to partition the image in to several small disjoint patches, while the region size, mean and variance features are used to calculate region cost for combination. Then in the second merging stage the DCT transform is used for energy compaction which is a criterion for texture comparison and region merging. Finally the image can be segmented into several partitions. The experimental results show that the proposed approach achieved very good segmentation robustness and efficiency, when compared to other state of the art image segmentation algorithms and human segmentation results.

Development of technique for face detection in image based on binarization, scaling and segmentation methods

“…-regions identification (regional growth, division and merging of regions, watershed) [9]; -histogram [10]; -based on partial differential equations [11]; -variation [12]; -graph [13]; -based on the Markov random field [14]. However, the issues related to improving the efficiency of the detected areas remained unresolved.…”

Improved fast mean shift algorithm for remote sensing image segmentation

“…Compared with pixel-based segmentation, the region-oriented method overcomes the heterogeneous noise that is inherent in pixel-based segmentation and extracts the features which are more meaningful and important for image interpretation. There exist several studies on object-oriented HR remote sensing image segmentation approaches [2][3][4]. It is evident that the HR images create additional challenges in terms of information extraction and classification [5][6][7].…”