Fig 2 - uploaded by Sebastian Houben
Content may be subject to copyright.
UK: Die Anzeige funktioniert lokal (Texmaker) richtig. Ich musste für Overleaf viele Nodes (Z.41-112) auskommentieren. Ich vermute, dass Overleaf mit vielen relativen Positionsberechnungen nicht zurecht kommt. Pipeline overview: For a given input image a license plate is detected, cropped and fed into the CycleGAN, which outputs a cartoon version of the input image as well as a binary license plate segmentation mask. The mask is used to determine the corners of the license plate, which are then adopted to render a new license plate template with a customized string, such as "ITSC 2021", in the same orientation on top of the cartoonized image. At the same time a second mask is derived from the detected edges, which is united with the original segmentation mask to achieve a more stable reinsertion process. The modified cartoon and mask are fed into the CycleGAN and the result is smoothly reinserted into the original image. Finally, a matching score between the two binary masks is calculated to serve as additional quality measure.
Source publication
Data augmentation techniques have been focused in recent research as they hold the promise to reduce the need for extensive data acquisition and to enable systematic sampling, e.g., in order to examine underrepresented cases. The question of how and to what extent control over the result is possible and necessary is still open. We propose a novel s...
Context in source publication
Context 1
... proposed pipeline for exchanging license plates in camera images is illustrated in Fig. 2. Initially, a pre-trained object detector [14] is used to localize license plates in images of vehicles in arbitrary poses. Each detected license plate is cropped and resized to match the input resolution of the CycleGAN, i.e., 256 × 256 pixels. The license plate crop is propagated through the cartoon generator of the CycleGAN, which ...
Similar publications
License Plate Recognition (LPR) is an important implemented application of Artificial Intelligence (AI) and deep learning in the past decades. However, due to the low image quality caused by the fast movement of vehicles and low-quality analogue cameras, many plate numbers cannot be recognised accurately by LPR models. To solve this issue, we propo...
