研究目的
To propose a novel method for salient object detection via double sparse representations under visual attention guidance, aiming to improve the object integrity and robustness for objects of different sizes.
研究成果
The proposed RSRVAG method performed better in the comparison with some state-of-the-art methods and was confirmed to be capable of working more effectively and efficiently on images with complex structures and detections of multiple objects. Future work should consider developing it by trying to weaken the dependence on initial visual attention guidance and enhance the integration strategies.
研究不足
The salient results obtained by sparse representation strictly rely on the dictionaries that were built simply based on the sorting of eye fixation results, which may fail when the visual attention guidance is poor. The integration of double sparse representation and multiscales is slightly simple and rough that may cause some under-detections or over-detections.
