<![CDATA[Logo detection is a challenging task in computer vision, especially when the logos are blurred or distorted in the images. Image deblurring is a technique that can improve the quality and clarity of the logos, which can enhance the logo detection performance. In this paper, we propose a novel method for logo detection that combines image deblurring and robust logo detection techniques. We create synthetic blurred images from the Flickr Logos 27 Dataset using Motion Blur data to improve deblurring methods. Then, we use three different image deblurring methods, namely Restormer, DeblurGAN-v2, and DeepRFT, to preprocess the images and remove the blur effects to improve the sharpness of images. We then use two different logo detection methods, namely Yolov7 and Robust Logo Detection, to detect and recognise the logos in the images. We evaluate our method on the Flickr Logos 27 dataset, which is a well-known and widely used dataset for logo detection. It contains 810 annotated images of 27 logo classes, as well as 4207 distractor images and 270 query images. We show that combining the method of Robust Logo Detection with Restormer achieves the highest mean average precision (mAP) at 0,754 among all the methods, and significantly improves the logo detection accuracy on blurred images. We conclude that image deblurring can effectively enhance logo detection performance and that our method is the best combination of image deblurring and logo detection techniques.]]>
Copyrights © 2024
