视觉显著性预测综述

Light field compression based on HEVC encoding and decoding

本文针对视觉显著性预测这一热点问题,通过介绍和对比了基于认知型、基于图论模型、基于频谱模型、和基于模式识别模型等视觉显著性预测算法的优缺点,得出基于模式识别模型在现有数据库效果最好的结论。同时,本文探究了已有的数据库特色以及适用范围。除此之外,本文分析了Auc、NSS和EMD等性能评估方法对于评估各种视觉显著性预测算法性能的优势和劣势。文章最后针对已有视觉显著性预测方法和数据库的问题,对现有的视觉显著性预测方面的研究做了总结和展望。

Based on the core problem of visualsaliency prediction, this paper firstly analyzed the imagefeature,then introduced and compared the advantages and disadvantages of the existing prediction methods, such asthese methods based on the graph model,cognitivemodel, patternclassification model and etcanddrewtheconclusionthatthepatternclassificationmethodsworkthebestamong existing dataset. Also, this paper summarized existing dataset, dataset characteristic and usable range for fixation prediction. Meanwhile, this paper used Auc, NSS and EMD performance evaluation methods to assess the performance of various visual fixation prediction algorithms.Finallythis papermade the summary and outlook on future work of visual saliency predictionespecially on existing methods and dataset.