以線散函數衍生特徵評估影像模糊度－應用於UAV影像篩選

無人飛行載具(Unmanned Aerial Vehicle, UAV)影像具備高解析及高重疊率等優勢，但輕巧的UAV載具容易受強風與紊流影響，載具的不穩定可能導致模糊影像的問題。傳統上影像品質指標可使用大型地面人工標計算點散函數(Point Spread Function, PSF)而得，本技術短文透過影像中人工構造物的線型特徵自動化評估線散函數(Line Spread Function, LSF)，首先使用直線段偵測(Line Segment Detector, LSD)演算法偵測影像中所有的直線段，其中階梯線(Step Edge)存在影像品質退化的趨勢，可應用於線散函數之評估。研究中以灰度值差條件篩選合格線，利用其邊緣散函數(Edge Spread Function, ESF)求得線散函數。最後藉由線散函數衍生特徵，由合格段之線散函數特徵擬合特徵橢圓，其特徵橢圓大小及長短軸比例可反映影像品質。研究資料為UAV影像及手持式GoPro Hero4序列影像，使用線散函數衍生特徵進行模糊影像分類實驗。本研究提出自動化評估影像品質的方法，並可進一步分類模糊影像，UAV影像及手持式影像之整體分類精度分別達88.9%及90%，驗證由線型特徵評估影像品質之可行性。

關鍵字

無人飛行載具；線散函數；線型特徵

並列摘要

Unmanned Aerial Vehicle (UAV) acquires high spatial resolution and highly overlapped images at low flying altitude. As the light-weight UAV is susceptible by strong winds and turbulence, the UAV instability will usually cause the blurred image and degrade the image quality. The image degradation function can be evaluated by point-spread function (PSF) and it is usually derived from the signalized target. This study presented an automatic LSF (line spread function)-derived feature to detect blurred image. It is based on the linear feature from the image itself. The linear features are detected by line segment detector (LSD) and the LSF is estimated by differentiating edge spread function (ESF). Only the step edges are preserved to calculate LSF-derived features such as size and azimuth of the ellipsoid. The test data includes UAV images and handheld images. We use the proposed LSF-derived features to separate the blur and non-blur images automatically. The overall accuracies reached 88.9% and 90% for UAV and handheld images, respectively. The experiments indicated that the proposed method is capable of detecting the blurred images automatically. Moreover, the blurred image can be excluded in photogrammetric processing to archive better accuracy.

並列關鍵字

Unmanned Aerial Vehicle ； Line Spread Function ； Linear Feature

國際替代計量

以線散函數衍生特徵評估影像模糊度－應用於UAV影像篩選

全文下載

主題瀏覽