研究目的
To propose an algorithm (dSLIC) that extends SLIC to dynamically adjust the local search, adopting superpixel resolution dynamically to structure existent in the image, providing more meaningful superpixels in the same linear runtime as standard SLIC.
研究成果
The dSLIC algorithm outperforms the SLIC solution in terms of undersegmentation error (20%) and achievable accuracy, providing more meaningful superpixels in the same linear runtime. Future work may include an extensive comparison and analysis against more works from the state-of-the-art.
研究不足
The limitations include the need for an initial segmentation and the computational load, albeit minimal, introduced by the dynamic adjustment of the search range.
1：Experimental Design and Method Selection:
The proposed method builds on SLIC, extending it to dynamically adjust the local search based on a structure measure.
2：Sample Selection and Data Sources:
Images from three datasets were used: The Berkeley Segmentation Dataset 500, The Stanford Background Dataset, and The Fashionista dataset.
3：List of Experimental Equipment and Materials:
An Intel Core i7 CPU at
4：40GHz-64GB was used. Experimental Procedures and Operational Workflow:
The dSLIC algorithm dynamically adjusts the search field size according to a structure measure, allowing for more meaningful superpixels.
5：Data Analysis Methods:
The performance was evaluated using undersegmentation error and achievable segmentation accuracy metrics.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
