| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 鄧名杉 | zh_TW |
| DC.creator | Ming-Sang Deng | en_US |
| dc.date.accessioned | 2014-7-29T07:39:07Z | |
| dc.date.available | 2014-7-29T07:39:07Z | |
| dc.date.issued | 2014 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=101522011 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 車牌影像的分析與應用是長久以來電腦視覺領域中一個重要的課題，在近年來，以此技術為基礎的應用正在蓬勃發展當中。舉例來說：道路監控、贓車追蹤、出入口監測等等，都是典型的應用。本篇論文除了一般狀況的車牌影像之外，更特別專注於透過超解析度技術來處理模糊車牌的案例。
本篇論文主要目標是設計一個能夠克服低解析度、光影變化，以及因為車距過遠所造成的車牌過小的狀況之影像還原系統。利用LBP特徵與RBF類神經網路進行特徵轉換，搭配影像分析演算法來進行模糊車牌的分析，接著利用兩階段式的影像還原，一步步將低解析影像還原成高解析影像，而每一次的還原都會產生一個新的影像，每一個新影像都會進行分析與投票。最終再依據投票結果進行影像重建的步驟。這個做法的目的在於透過一次又一次的分析與還原的過程中，強化原本因為模糊而弱化的影像特徵。
在實驗方面分成三大部分。第一部分採用模糊但肉眼可辨識的車牌影像，其目標為還原出正確且易讀的車牌。第二與第三部分則採用模糊且肉眼無法辨識的車牌影像，並透過「候選名單」的排名，來進行準確度的分析。該候選名單可以用來做為模糊車牌解答之參考依據，藉以降低模糊車牌比對上的搜尋範圍。實驗結果顯示，本論文所提出的方法對於模糊車牌有很高的還原成功率，而即便在非常模糊以至於肉眼無法分辨的車牌影像上，仍然可以有很好的辨識成效。
| zh_TW |
| dc.description.abstract | The research on vehicle license plate is an important issue in computer vision. In recent years, applications based on this technology are getting popular. For example, entrance monitoring system, road surveillance, suspicious vehicle investigation…etc. In our work, we are not only dealing with the normal case but also focusing on the blurred plate images using super-resolution technique. The purpose of this thesis is to design a vehicle license plate analysis and image reconstruction system that can overcome the following cases: blurred images, images with variation illumination intensity, and tiny target images.
In this thesis, we adopt LBP as image feature and RBF neural network to perform feature translation in the first stage. After that, an algorithm for analyzing the image and performing the image reconstruction is proposed in the second stage. In image reconstruction stage, two-layer architecture is applied. In the first step, image reconstruction is performed recursively to obtain better reconstruction result. A new image will be generated each time and each new image will go through analysis and voting procedure recursively. In the second step, image reconstruction is further manipulated according to the voting result. The purpose of image reconstruction is to intensify the weak clue in the blurred image recursively.
Three different experiments were conducted to verify the validity of our proposed method. The images in the first experiment are blurred but can be identified by human. We can apply the proposed method to reconstruct the original blurred images into clear and correct license plate images. The images of the second and third experiments are very blurred and cannot be identified by human. Here, we utilize the candidate ranking list for analyzing the accuracy. This list can help police to reduce the search space when performing the blurred license plate image matching for criminal investigation.
| en_US |
| DC.subject | 車牌切割 | zh_TW |
| DC.subject | 車牌辨識 | zh_TW |
| DC.subject | 超解析度 | zh_TW |
| DC.subject | 影像還原 | zh_TW |
| DC.subject | Vehicle license plate segmentation | en_US |
| DC.subject | vehicle license plate recognition | en_US |
| DC.subject | super-resolution | en_US |
| DC.subject | image reconstruction | en_US |
| DC.title | 模糊車牌字元預測使用超解析度影像重建技術 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Blur License Plate Character Prediction Using Super-Resolution Based Image Reconstruction Technique | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |