CAJ | 학술논문

目的一幅图像或场景的显著性区域代表它们的主要内容(显著目标).由于视觉假体可植入电极的数量有限,只有低分辨率的图像对其才有用,因此提取显著性区域有助于视觉假体捕捉到场景中的显著目标.方法 Itti模型是一个显著性检测模型,它检测到的显著性区域与人的视觉感知有差异,显著目标的边界不明确.笔者去除了Itti模型提取的方向和色彩特征,将红(R)、绿(G)、蓝(B)三基色(RGB)图像转换到对应于HSI颜色空间上的色调(H)、纯度(S)、亮度(I)3个新特征分量,对hti模型进行优化改进.在显著图中,将落在显著目标内的显著点面积与总显著点面积的比值定义为显著图精确度;以显著图精确度为提取显著图方法的测度,对改进前后2种方法进行比较.结果利用改进方法提取的显著图比Itti模型显著图精确度提高了约20％;在检测显著性区域时所用时间减少近50％.结论提出了一种用于人工视觉的获取显著目标的方法,本算法可以得到更加精确的显著性结果,且可缩短运行时间.
목적 일폭도상혹장경적현저성구역대표타문적주요내용(현저목표).유우시각가체가식입전겁적수량유한,지유저분변솔적도상대기재유용,인차제취현저성구역유조우시각가체포착도장경중적현저목표.방법 Itti모형시일개현저성검측모형,타검측도적현저성구역여인적시각감지유차이,현저목표적변계불명학.필자거제료Itti모형제취적방향화색채특정,장홍(R)、록(G)、람(B)삼기색(RGB)도상전환도대응우HSI안색공간상적색조(H)、순도(S)、량도(I)3개신특정분량,대hti모형진행우화개진.재현저도중,장락재현저목표내적현저점면적여총현저점면적적비치정의위현저도정학도;이현저도정학도위제취현저도방법적측도,대개진전후2충방법진행비교.결과 이용개진방법제취적현저도비Itti모형현저도정학도제고료약20％;재검측현저성구역시소용시간감소근50％.결론 제출료일충용우인공시각적획취현저목표적방법,본산법가이득도경가정학적현저성결과,차가축단운행시간.
Objective Salient regions may express the main content (salient objects) in an image or scene.Since the number of implanted electrodes for visual prosthesis is limited,only the low-resolution image is used.Extraction of the salient regions may help capture salient objects in the scenes of visual prosthesis.Methods Salient regions detected by Itti model,one of salient detection models,have some differences with human visual perceptions and its salient objects boundary are also unclear.The Itti model was optimized by removing the orientation and the color characteristics,and converting the RGB image in an image or scene into the HSI color space with three new feature components of hue,saturation and intensity.In saliency map,the ratio of the salient point area in salient objects regions and the total salient point area was defined as salient accuracy and computed,which was used to compare the effects of the two methods of extracting salient map.Results Compared with the Itti method for extracting saliency map,the salient accuracy of the improved method was increased by about 20％,and the time for detecting the same salient regions by our method was shorter by nearly 50％.Conclusions The proposed algorithm can be an alternative method to extract salient map from a complex image or scene for retinal prosthesis with higher salient accuracy and less running time.