基於學習方法及分類基底重建車牌影像的超解析度技術

Example-based License Plate Image Super-Resolution Using Classified Bases

許穎華(Ying-Hua Hsu)

指導教授 ： 林嘉文

國立清華大學/電機資訊學院/電機工程學系/碩士(2011年)

摘要

數位攝影機將拍攝而成的影像儲存在數位硬碟裡面，為了能夠更有效率的儲存更多的資訊，影像品質就必須要被選擇性的壓縮。壓縮後的影像，畫面品質較難一眼就清楚地看出細節部份，只能夠觀察到場景的大架構，這對於必須從影片細節中得取資訊的使用者來說，無疑是一種阻礙，就像是我們很難由一張影像品質粗劣的車牌影像中去直接辨識出車牌號碼。為了從影像品質惡劣的車牌影像中去得到更多的車牌資訊，我們針對了這個主題去做更進一步的研究。本研究利用分類基底與基於學習方法的超解析度技術由原始低解析度的車牌影像去重建高解析度的車牌影像，分類基底對應到事先分好類別的各類樣板，而基於學習方法的超解析度技術則再使用與原始車牌字元影像相似類別中的基底去重建高解析度車牌影像。先將車牌影像中的單一字元切割出並投影至Orthogonal Locality Preserving Projections (OLPP) 空間中，再使用貝氏分類器 (Bayesian classifier) 找出所屬的字元類別， 從定義的cost function訂正由貝氏分類器分類後的錯誤結果，最後再經由類別判定的步驟得到每個字元最有可能的分類。由分類的結果與事前訓練的合成基底，我們可以使用與每個字元最合適的基底集合去做字元影像的重建。最後將各個字元重建的結果合成為一張高解析度的車牌影像。

關鍵字

超解析度

並列摘要

It is hard to distinguish the character from the car license plate image which has poor quality. We propose a method using classified bases and example-based super-resolution to reconstruct a high-resolution car license plate image. We use the Bayesian classifier to classify the single character of the license plate in Orthogonal Locality Preserving Projections (OLPP) subspace. The proposed cost function in the refinement step correct the classified result in the following. According to the classified result of final class decision step, we can select the most suitable basis set to reconstruct a high-resolution car license plate image. The experimental results can show that the reconstructed image via our proposed method is better than the result which reconstructed by calculating the whole car license plate image.

並列關鍵字

Super-Resolution

參考文獻

[1] Y. Jie, D. Si-dan, and Z. Xiang, “Fast Super-resolution for License Plate Image Reconstruction”, ICPR, 2008.

[2] K.V. Suresh, G.M. Kumar, and A.N. Rajagopalan, “Superresolution of License Plates in Real Traffic Videos”, IEEE Trans. Intelligent transportation systems, June 2007, Vol. 8, No. 2.

[3] S. Vasuhi and V. Vaidehi, “Identification of Human Faces using Orthogonal Locality Preserving Projections”, International Conference on Signal Processing Systems, 2009.

[4] A. M. Martinez and A. C. Kak, “PCA versus LDA”, IEEE Trans. Pattern Anal. Mach. Intell., vol. 23, no. 2, pp. 228-233, Feb. 2001.

[6] M. Elad and A. Feuer, “Restoration of a single superresolution image from several blurred, noisy, and undersampled measured images,” IEEE Trans. Image Processing, vol. 6, no. 12, pp. 1646-1658, Dec. 1997.