陳浩 郭欣欣
摘 ? 要:圖像插值是從已知像素值計(jì)算未知像素值的過程,可用于數(shù)字圖像的放大和對(duì)比。圖像插值技術(shù)在遙感、醫(yī)學(xué)診斷、農(nóng)業(yè)、地質(zhì)、軍事等眾多領(lǐng)域都有著廣泛的應(yīng)用。有多種算法可用于圖像縮放,這些技術(shù)主要分為自適應(yīng)和非自適應(yīng)圖像插值兩類,這兩種插值技術(shù)可進(jìn)一步分為各種類型。本文從峰值信噪比(PSNR)這一性能參數(shù)出發(fā),對(duì)這些不同的自適應(yīng)和非自適應(yīng)圖像插值技術(shù)進(jìn)行了闡述和比較。選擇合適的插值方法是一項(xiàng)非常嚴(yán)謹(jǐn)?shù)难芯浚彩沁M(jìn)行插值分析的首要要求。
關(guān)鍵詞:圖像插值 ?非自適應(yīng) ?自適應(yīng) ?峰值信噪比
中圖分類號(hào):TP391 ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 文獻(xiàn)標(biāo)識(shí)碼:A ? ? ? ? ? ? ? ? ? ? ? ?文章編號(hào):1674-098X(2020)04(c)-0128-03
Abstract: Image interpolation is the process of calculating unknown pixel values from known pixel values, and can be used to zoom in and compare digital images. Image interpolation technology has been widely used in many fields such as remote sensing, medical diagnosis, agriculture, geology, and military. There are a variety of algorithms available for image scaling. These techniques are mainly divided into adaptive and non-adaptive image interpolation. These two interpolation techniques can be further divided into various types. Based on the performance parameter of PSNR, this paper describes and compares these different adaptive and non-adaptive image interpolation algorithms. Choosing a suitable interpolation method is a very rigorous study and the first requirement for interpolation analysis.
Key Words: Image interpolation; Non-adaptive; Adaptive; PSNR
圖像插值在圖像處理領(lǐng)域有著廣泛的應(yīng)用,比如可以應(yīng)用于計(jì)算機(jī)圖形學(xué)、繪制、編輯、醫(yī)學(xué)圖像構(gòu)建和輪廓圖像查看等各個(gè)領(lǐng)域。圖像插值就是在不丟失原始圖像的視覺信息的情況下,從原始低分辨率圖像分辨率生成另外一個(gè)高分辨率分辨率的圖像[1]。圖像縮放是將圖像從一個(gè)比例轉(zhuǎn)換為另一個(gè)比例的過程,而插值是一個(gè)圖像放大的過程,這個(gè)過程是通過在離散輸入樣本區(qū)間的值擬合一個(gè)連續(xù)函數(shù)來實(shí)現(xiàn)[2]。圖像放大技術(shù)已廣泛應(yīng)用于計(jì)算機(jī)設(shè)備、打印機(jī)、數(shù)字電視、媒體播放器等領(lǐng)域。另一方面,圖像縮小技術(shù)在應(yīng)用中也很有用,例如將高分辨率圖像轉(zhuǎn)換為低分辨率圖像,以適應(yīng)小型液晶顯示系統(tǒng)。圖像插值是數(shù)字圖像處理中一個(gè)具有挑戰(zhàn)性的重要問題。
插值算法可分為自適應(yīng)和非自適應(yīng)兩類。非自適應(yīng)算法將固定模式應(yīng)用于每個(gè)像素,而不考慮其他參數(shù)作為圖像的特征、邊緣。這種插值技術(shù)可能會(huì)在插值圖像中產(chǎn)生鋸齒效應(yīng)等[2]。自適應(yīng)算法利用鄰域像素的光譜和空間特征,使未知像素盡可能接近原始像素。為了獲得高質(zhì)量的圖像,人們開發(fā)了空間和光譜相關(guān)技術(shù)。根據(jù)相鄰像素的特征,預(yù)測新像素的值,使其更有效地插值。在自適應(yīng)插值技術(shù)中減少或消除了鋸齒效應(yīng)。但是,一些自適應(yīng)插值技術(shù)可能需要更多的時(shí)間來轉(zhuǎn)換圖像,因此有時(shí)它可能無法應(yīng)用于實(shí)時(shí)應(yīng)用。插值方法有兩空域插值和頻域插值[3]兩種。近年來,空間域技術(shù)因其復(fù)雜度低而被應(yīng)用于實(shí)時(shí)應(yīng)用中。另一種方法使用各種變換,如離散余弦變換(DCT)、離散傅立葉變換(DFT)或小波變換在頻域中縮放圖像。頻域技術(shù)獲得了更高的復(fù)雜度和存儲(chǔ)需求,因此不適合實(shí)時(shí)或低成本的應(yīng)用。使用現(xiàn)場可編程門陣列(FPGA)或?qū)S眉呻娐罚ˋSIC)可以實(shí)現(xiàn)這些插值技術(shù)[2]。
1 ?非自適應(yīng)圖像插值算法
在各種提出的實(shí)現(xiàn)圖像縮放的非自適應(yīng)插值算法中,主要有最近鄰、雙線性、雙三次圖像插值等。最簡單的是最近鄰算法,時(shí)間復(fù)雜度低,并且比較容易實(shí)現(xiàn)[4]。使用此算法創(chuàng)建的圖像包含塊化和鋸齒化算法,為了減少這種阻塞和混疊效應(yīng),最常用的方法是雙線性算法,該算法使用線性插值模型來計(jì)算未知像素。雙三次插值是比較復(fù)雜但更精確的方法,通過二維規(guī)則網(wǎng)格的加權(quán)和插值圖像像素,可以得到高質(zhì)量的圖像[5]。
1.1 最近鄰插值
最近鄰插值是最簡單的插值算法,因?yàn)橹豢紤]一個(gè)最接近插值點(diǎn)的像素,所以在所有插值算法中,這種算法處理時(shí)間最短,但這會(huì)使每個(gè)像素變大。在圖像像素分辨率過高的情況下,該算法具有很好的效果。最近鄰插值的插值核為
自適應(yīng)插值方法在圖像視覺質(zhì)量方面得到了提升,但它需要更多的計(jì)算投入。當(dāng)時(shí)間不是決定性因素時(shí),我們一般選擇自適應(yīng)策略,但非自適應(yīng)策略更簡單。所以,在實(shí)際應(yīng)用中,根據(jù)不同的圖像處理對(duì)象,選擇不同的插值算法。
參考文獻(xiàn)
[1] Mahajan S H , Harpale V K . Adaptive and Non-adaptive Image Interpolation Techniques [C]// Computing Communication Control and Automation Conference(ICCUBEA), 2015 IEEE. IEEE, 2015.
[2] Jaiswal S P , Jakhetiya V , Kumar A , et al. A low complex context adaptive image interpolation algorithm for real-time applications[C]// Instrumentation and Measurement Technology Conference (I2MTC), 2012 IEEE. IEEE, 2012.
[3] 高翔. 基于DCT域的圖像插值技術(shù)研究[D].哈爾濱工業(yè)大學(xué),2012.
[4] Patel V, Mistree K. A Review on Different Image Interpolation Techniques for Image Enhancement[J]. ?International Journal of Emerging Technology and Advanced Engineering, 2013, 3(12):129-133
[5] Yunshan Zhang, Yuhui Li, Jie Zhen, Jionghao Li, Ran Xie. The Hardware Realization of the Bicubic Interpolation Enlargement Algorithm Based on FPGA[P]. Information Processing (ISIP), 2010 Third International Symposium on,2010.
[6] Wang Xiuhua,Jia Xinyue,Zhou Wei,Qin Xiaoyun,Guo Hanming. Correction for color artifacts using the RGB intersection and the weighted bilinear interpolation.[J]. Applied optics,2019,58(29).
[7] Lee J H , Kim J O , Han J W , et al. Edge-Oriented Two-Step Interpolation Based on Training Set[J]. IEEE Transactions on Consumer Electronics, 2010, 56(3):1848-1855.
[8] Chen S L . VLSI Implementation of an Adaptive Edge-Enhanced Image Scalar for Real-Time Multimedia Applications[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2013, 23(9):1510-1522.
[9] Karpagam A V , Manikandan M . Interpolation tool: Qualitative and quantitative analysis for machine vision applications[C]// Fourth International Conference on Signal Processing. IEEE, 2017.
[10]Hwamog Kim, Jeffrey L. Willers, Seongjai Kim. Curvature Interpolation Method for Surface Reconstruction for Geospatial Point Cloud Data[J]. International Journal of Remote Sensing, 2019:1-30.
[11]房琰. 基于梯度輪廓和非局部自相似性的單幅圖像超分辨率方法研究[D].山東大學(xué),2019.
[12]Zhu S, Zeng B, Zeng L, et al. Image Interpolation Based on Non-local Geometric Similarities and Directional Gradients[J]. IEEE Transactions on Multimedia, 2016, 18(9):1-1.
[13]Dong W, Shi G, Xu J. Adaptive Nonseparable Interpolation for Image Compression With Directional Wavelet Transform[J]. IEEE Signal Processing Letters, 2008(15):233-236.
[14]汪成峰. 基于自適應(yīng)關(guān)節(jié)權(quán)重和插值小波的體感動(dòng)作評(píng)價(jià)方法研究[D].中國農(nóng)業(yè)大學(xué),2016.
[15]李小燕. 基于廣義圖像先驗(yàn)的圖像超分辨率重建算法研究[D].西南交通大學(xué),2015.
[16]Liu X , Zhao D , Xiong R , et al. Image Interpolation Via Regularized Local Linear Regression[J]. IEEE Transactions on Image Processing, 2012, 20(12):3455-3469.