Integration of Image-Derived and Pos-Derived Features for Image Blur Detection

Abstract

Commission I, ICWG I/VbKEY WORDS: Position and Orientation System (POS), Unmanned Aerial Vehicle (UAV), blur image detection, degree of motion blur ABSTRACT:The image quality plays an important role for Unmanned Aerial Vehicle (UAV)'s applications. The small fixed wings UAV is suffering from the image blur due to the crosswind and the turbulence. Position and Orientation System (POS), which provides the position and orientation information, is installed onto an UAV to enable acquisition of UAV trajectory. It can be used to calculate the positional and angular velocities when the camera shutter is open. This study proposes a POS-assisted method to detect the blur image. The major steps include feature extraction, blur image detection and verification. In feature extraction, this study extracts different features from images and POS. The image-derived features include mean and standard deviation of image gradient. For POS-derived features, we modify the traditional degree-of-linear-blur (b linear ) method to degree-of-motion-blur (b motion ) based on the collinear condition equations and POS parameters. Besides, POS parameters such as positional and angular velocities are also adopted as POS-derived features. In blur detection, this study uses Support Vector Machines (SVM) classifier and extracted features (i.e. image information, POS data, b linear and b motion ) to separate blur and sharp UAV images. The experiment utilizes SenseFly eBee UAV system. The number of image is 129. In blur image detection, we use the proposed degree-of-motion-blur and other image features to classify the blur image and sharp images. The classification result shows that the overall accuracy using image features is only 56%. The integration of image-derived and POS-derived features have improved the overall accuracy from 56% to 76% in blur detection. Besides, this study indicates that the performance of the proposed degree-of-motion-blur is better than the traditional degree-of-linear-blur.

Fusion network for blur discrimination

“…Teo and Zhan 15 Liu et al 14 Xu et al 11 Rugna and Konik 13 Huang et al 18 Ours It is found that accuracy of these datasets is all greater than 94%, which again demonstrates our method has superior generalization ability and satisfactory robustness. Considering blurry images and sharp images separately, all precision and recall are greater than 94% except GoPro dataset.…”

“…Liu et al 14 focused their attention on low-level features and proposed a blur discrimination method through some image features, including local auto-correlation congruency, gradient histogram span, spectrum slope, and maximum saturation. Teo and Zhan 15 proposed a detection method for the blurry image by integrating image-derived features and position and orientation system-derived features. Wang et al 16 proposed a blur detection method for iris image based on local features, which are generated by radial symmetry transform and support vector machine.…”

“…In the last 20 years or so, several practicable methods are proposed for blur discrimination. [10][11][12][13][14][15][16] Most of them employ a two-step strategy. First, some low-level blur-related features are handcrafted based on various empirical image statistics in gradient, frequency, and other domains.…”