應用於高解析度影像系統處理之FPGA IP設計

IPs Design of FPGA for High-resolution Image System Process

王星堡(Xing-Bao Wang)

指導教授 ： 黃文增 ； 譚巽言

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2013年)

摘要

隨著科技進步，監控系統的影像解析度已達Full-HD(Full High Definition，1920x1080，1080p)格式，而影像更新頻率為超過30 fps(Frames Per Second)之規格。為了達到Full-HD的規格並擁有更好的影像品質，本研究應用FPGA作為高解析度影像系統處理(ISP，Image System Process)核心，配合類比前端模組與CCD影像感測器的特性，設計一套可支援4種不同解析度的影像處理流程。本研究ISP流程主要包含8個知識產權(Intellectual Properties，IP)核心，自動曝光、自動白平衡、色彩濾波陣列內插法、色彩校正、雜訊濾波器、色彩空間轉換、動態範圍校正、以及伽瑪校正。藉由對4種不同解析度感測器的支援，8個可開關的影像處理步驟，和HDMI、BT1120、Raw Data三種不同的輸出格式，本系統可以應用到不同平台與輸出客製化的資料格式，做為各種領域的影像來源。
本研究是應用影像的SNR、白平衡誤差，和TVL(TeleVision Lines)來評估影像的品質。在此，我們以2百萬像素(簡稱2M)感測器當做量測裝置，在相同環境和不同感光度(ISO)下和商品化數位相機比較SNR效能，本研究的效能都較佳。更進一步，接著在燈箱內以不同色溫測量，我們的白平衡誤差約為0.02(HSV saturation)，此誤差已經是讓人眼難以察覺2種不同顏色之程度。TVL則為1361 LW/PH (Line Width/Picture Height)，約為理想值的71%。因此，以上結果證明本研究的ISP系統是可以當成一個可靠的影像來源。

關鍵字

影像系統處理 ； 高解析度攝影機 ； CCD ； SNR ； 白平衡誤差 ； 知識產權

並列摘要

With the progress of science and technology, the surveillance image resolution system is upto Full-HD (Full High Definition, 1920x1080 resolution, 1080p), and its frame rate is over 30 fps (Frames Per Second). In order to achieve Full-HD specifications and get a better image quality, this study uses FPGA as ISP (Image System Process), with the characteristics of the CCD image sensor and analog front-end module, designed an image processing pipeline which can support of four different resolutions. The image processing procedure included eight major IP (Intellectual Properties) cores, auto exposure, auto white balance, color filter array interpolation, color correction, noise filter, color space conversion, dynamic range correction, and gamma correction. With supporting the 4 different resolutions sensors, eight switchable image processing steps, and three different output formats (HDMI, BT1120, and Raw Data), our system can applied to different platforms and outputted the customized data format as the image source of various fields.
The 2 megapixels (briefly called 2M) sensor is as the measurement device in this study. Compared with the SNR performance of the commercialized CCD digital cameras, our study has the better performance than others under the same tested parameters and different sensitivities (ISO). Moreover, under the tested light boxes with different color temperature measurements, the white balance error of this study is about 0.02 (HSV Saturation) to be difficultly perceived the degree of the difference 2 colors by human eyes. TVL is 1361 LW/PH (Line Width/Picture Height), which is approximately 71% of the ideal condition. Therefore, our ISP system can be proved as a reliable video source from above results.

並列關鍵字

Image System Process ； High resolution camera ； CCD ； SNR ； White balance error ； Intellectual Property

參考文獻

[12] 林殷旭，自動白平衡之軟硬體共同設計，碩士論文，國立台北科技大學電子工程系，2006。

[32] 羅一中，改良式直方圖等化應用於即時影像增強之低成本FPGA設計，碩士論文，台北科技大學電機工程學系，2005。

[4] W. C. Kao, “Real-time Image Fusion and Adaptive Exposure Control for Smart Surveillance Systems,” Electronics Letters, Vol. 43, No. 18, pp. 975-976, 2007.

[5] T. Jiang, K. D. Kuhnert, D. Nguyen, and L Kuhnert, “Multiple templates auto exposure control based on luminance histogram for onboard camera,” IEEE International Conference on Computer Science and Automation Engineering (CSAE), Vol. 3, pp. 237-241, 2011.

[6] L. JiaYi, Q. YaJie, and H. Zhiliang, “An auto-exposure algorithm for detecting high contrast lighting conditions,” 7th IEEE Conference on ASIC, pp. 725-728, 2008.