关键词:道路模型 感知编组 动态规划 道路提取 知识推理 图像理解 低分辨率 遥感影像 道路网 自动提取 方法 研究 Remote Sensing Image Resolution Automatic Extraction Roads 普适性 识别过程 道路跟踪算法 SPOT 卫星 传感器 优势 计算效率 人工干预 全自动
Image understanding is generally defined as the construction of explicit, meaningful descriptions of the structure and the properties of the 3-dmi ensionalworld from 2-dmi ensional mi ages. A conceptual framework for mi age understanding is based onMarr’s concept of visual perception as computational process. Marr postulated a hierarchical architecture for vision systemswith different intermediate representations and processing levels ( low, middle and higher level vision). According to the description ofMarr’smachine vision theory and the characteristicsof low resolution remote sensing mi ages, this paperproposes an automaticmain-road extractionmethod, which isbased on line segmentperceptual grouping and dynamic programming. Themethod exploits the road model in low resolution remote sensing mi ages at low level and locates the road seeds automatically by line segmentperceptualgrouping atmiddle leveland then tracks the road seeds to extract the road networks by dynamic programming athigh leve.l Firstlywe illustrate the roadmodel in low resolution remote sensing mi ages based on analyzing road characteristics,such as photometric, geometric, topological and contextual characteristics and so on. To increase the precision of road object recognition and to reduce the effects of noise, source mi ages are preprocessed ahead, which include contrast stretching, edge information detection with canny operator and redundant line segments elmi ination. Edge detection is crucial to line segmentperceptual grouping, thus canny operator is applied because of its characteristics ofhigh position precision, single pixelwidth and low error rate. In the process of line segments elmi ination, the length and curvature of line segments are considered as the decisive factors to the elmi ination ofredundant line segments. But the directionsof the beginning and end line segments are recorded to assist the decision. Atmiddle leve,l edge line segments are grouped by perceptual grouping technology based on contextual line segments to generate latent road edge line segments. After that,several road seeds are located by computing the latent road edge line segment groups. The relationship of gray value between the regions shaped by the latent road edge line segmentand theirbackground is exploited to locate the road seeds thereinto. Then a new road tracking approach using dynamic programming is adopted at high leve.l The approach introduces the conceptofminmi ized cost route and extends the prmi itive segmentwhich is formed by direct connection of\nroad seeds to thewhole road network in light ofminmi ized cost route. Finally false alarms are elmi inated by knowledge inferencemethod, inwhich inference rules are attained based on the geographic characteristics of road networks in low resolution remote sensing mi ages.We conducted expermi ents on three datasets of low resolution remote sensing mi ages, which include the Landsat7 (B80 band) mi agewith15m-resolution, the SPOT mi age ofSanDiego districtwith 10m-resolution and the SAR(Synthetic ApertureRadar) mi agewith 12·5m-resolution. Correctness and completeness are introduced tomake objective evaluation on the effectivenessof themethod. In thisway, reference data, whichmeans the road network plotted by observer, should be defined ahead. Expermi ental results show thatourproposedmethod hashigh correctness, especially inLandsat7 remote sensing mi age(98·7% ). Meanwhile the completeness criteria gained from all the source datasets is comparatively high.The lowestvalue(88·1% ) appears in SAR mi age probably due to the speckle noises. Moreover the followings are proved by the expermi ental results: (1) the proposedmethod is effective in lowresolution remote sensing mi ages (high resolution remote sensing mi ages can be sampled to generate its low resolution counterpart). Especially the mi ages contain some sparse rural roads and intricate city road network; (2) themethod is completely automatic and shows better computation efficiency than others, especially compared to semi-automatic road detection methodswhich need human and computer interaction; (3) themethod shows robustness and good performance in remote sensing mi ages such as Landsat7, SPOT and SAR.