摘要
提出一种基于小波并结合生物视觉机制的感兴趣区域提取算法,首先使用小波构建亮度、颜色和方向等3个特征的小波金字塔,然后使用中央—周边算法得到每个特征的关注图,对关注图进行迭代运算以进一步抑制不显著区域。对迭代处理后的显著图进行线性组合得到最终的显著图;使用WTA网络和返回抑制机制提取显著点,以显著点为中心进行区域生长得到最终的显著区域。实验证明,本方法在提取效果上较原方法提高了4.2%和效率上较原方法提高了28.75%,总体上较原方法有了明显改善。
An extraction algorithm was presented based on wavelet and biological vision mechanism.First,we used wavelet to construct a wavelet pyramid with characteristics of brightness,colour and direction,then employed centeral-circumjacent algorithm to obtain attention map of each characteristic,and carried out iterative algorithm to restrain non-salient regions.Finally,we linearly combined the attention maps to get final attention map and used WTA network and inhibitory mechanism of return to extract salient point,using which as the center to get ultimate saliency region.The test shows that this approach is improved on extraction result and efficiency by 4.2%,and 28.75% respectively,compared with the previous method generally.
出处
《杨凌职业技术学院学报》
2011年第1期13-15,共3页
Journal of Yangling Vocational & Technical College
关键词
图像检索
小波
感兴趣区域
显著图
image retrieval
wavelet
region of interest
saliency map
