摘要
在视觉注意机制下,图像显著性表现为人眼对图像中部分区域的重视程度,但采用频率调谐(frequency tuning,FT)算法进行图像显著性检测时,出现显著区域亮度不均匀、细节信息不明显等问题。针对上述问题,提出一种基于对比度增强的图像显著性检测方法。该方法将输入图像进行全分辨率对比度增强处理,通过高斯带通滤波器筛选出显著性区域,然后将Lab颜色空间分解成3个特征通道并做归一化处理,利用各通道的能量占比进行动态权值融合得到最终显著图。结果表明,该方法提高了目标轮廓及内部细节信息的清晰度,明显地突出显著区域,更易于目标的检测与识别。
Under the visual attention mechanism,image saliency presents the degree that human eyes attach importance to some regions of the image.However,when the frequency tuning(FT)algorithm is used to detect the image significance,some problems occur,such as uneven brightness of the significant area and unclear details.Hence,a method for image saliency detection based on contrast enhancement is introduced.In this method,the contrast of input image was full-resolution enhanced,and salient regions were extracted via Gaussian band-pass filter,then the Lab color space was decomposed into three feature channels and normalized.According to the power of each channel,dynamic weight was distributed to combine into a final saliency map.The results show that the clarity of the target profile and internal details are improved,and the salient region is highlighted clearly,which is much easier for target detection and recognition.
作者
彭明
杨延竹
龚嘉杰
PENG Ming;YANG Yanzhu;GONG Jiajie(College of Mechanical Engineering,Donghua University,Shanghai 201620,China)
出处
《东华大学学报（自然科学版）》
CAS
北大核心
2019年第6期888-891,917,共5页
Journal of Donghua University(Natural Science)
关键词
图像显著性检测
对比度增强
频率调谐
动态权值
image saliency detection
contrast enhancement
frequency tuning
dynamic weight
