Classifying Discriminative Features for Blur Detection

Abstract

Blur detection in a single image is challenging especially when the blur is spatially-varying. Developing discriminative blur features is an open problem. In this paper, we propose a new kernel-specific feature vector consisting of the information of a blur kernel and the information of an image patch. Specifically, the kernel specific-feature is composed of the multiplication of the variance of filtered kernel and the variance of filtered patch gradients. The feature origins from a blur-classification theorem and its discrimination can also be intuitively explained. To make the kernel-specific features useful for real applications, we build a pool of kernels consisting of motion-blur kernels, defocus-blur (out-of-focus) kernels, and their combinations. By extracting such features followed by the classifiers, the proposed algorithm outperforms the state-of-the-art blur detection method. Experimental results on public databases demonstrate the effectiveness of the proposed method.

Motion Blur Detection With an Indicator Function for Surveillance Machines

“…Table III shows that the proposed WSV and WSV+Grabcut have significantly higher detection accuracy than DBDF. In addition, WSV outperforms MVV and WSV+Grabcut outperforms MVV+Grabucut in terms of accuracy, implying that the proposed one-class classifier is superior to the non-linear SVM employed in MVV and MVV+Grabcut [29]. It can also be seen from Table III that WSV is one order faster than MVV and two orders faster than DBDF.…”

“…The proposed WSV algorithm is compared with LBM [14], [32], DBDF [15], and MVV [29]. Though there are several advanced blur algorithms [15], they employ much richer features.…”

“…However, this indicator function can only tell a region is blurred by a predefined kernel k or not blurred at all (i.e., the point spread function is the impulse function h ). To discriminate between k and any kernelk with k = h andk = k, the MVV method [29] The description of the proposed method is graphically supported by Fig. 3 where the vectorsw and d are used whereasṽ is discarded.…”

Self-supervised blur detection from synthetically blurred scenes

“…A number of previous works have investigated blur localization directly or implicitly from the feature engineering and physical modeling approaches, either taking one single [12,11,3] or multiple [13,14,15] images as input, and aiming at detecting only one [16,12,6,17,18,19,20] or both kinds of blur [21,22,23]. Most of these try to leverage information extracted directly from the intensities [21], from the gradients [24,23,25,20], or from transformed domains [22,16,12,11,26].…”

“…More recently, supervised learning-based approaches [25], and particularly those based on the use of Convolutional Neural Networks (CNN), have shown enormous potential for tackling tasks that require a dense, per-pixel prediction, such as semantic segmentation [27,28], instance segmentation [29] or crowd counting via density map estimation [30]. Blur segmentation can also be viewed as one of such dense prediction tasks, and several works have already explored this approach, either for predicting both types [31,1,32] or defocus only blur [33,34,35,36].…”