Удаление шума из изображения с использованием разреженного представления (1187432), страница 4
Текст из файла (страница 4)
Ïðåäñòàâëÿåòñÿ áîëåå åñòåñòâåííûì îáðàòíûéïîðÿäîê ãåíåðàöèè, êîòîðûé è èñïîëüçóåòñÿ â ïðåäëîæåííîé âåðîÿòíîñòíîé ìîäåëè.Ñðàâíèì äàëåå ïðåäëîæåííûé ìåòîä ñ ìåòîäîì CSR è ìåòîäîì K-SVD íà ïàðåèçîáðàæåíèé ñî ñòàíäàðòíûì îòêëîíåíèåì øóìàσ = 20èσ = 30(ñì. ðèñ.23).Çíà÷åíèÿ ñîîòíîøåíèÿ PSNR äëÿ âñåõ òðåõ ìåòîäîâ ïðèâåäåíî â òàáë. 1.Òàáëèöà 1: Ñðàâíåíèå ïî PSNR ïðåäëîæåííîãî ìåòîäà óäàëåíèÿ øóìà ñ ìåòîäàìèCSR è K-SVDσ = 20σ = 30K-SVDCSRÏðåäëîæåííûé ìåòîä31.6331.7831.7929.7929.7329.75Èç òàáë. 1 çàêëþ÷àåì, ÷òî ïðåäëîæåííûé ìåòîä èìååò ñõîäíîå êà÷åñòâî óäàëåíèÿ øóìà ïî îòíîøåíèþ ê ìåòîäó CSR, à ïîòîìó ïðåäñòàâëÿåòñÿ ïåðñïåêòèâíûìäëÿ ðåøåíèÿ çàäà÷è óäàëåíèÿ øóìà.
 äàëüíåéøåì â ðàìêàõ ïðåäëîæåííîé ñòàòèñòè÷åñêîé ìîäåëè ïëàíèðóåòñÿ èçó÷èòü äðóãèå àïðèîðíûå ðàñïðåäåëåíèÿ, êîòîðûåèìåþò áîëåå ñèëüíûå ñâîéñòâà ïîîùðåíèÿ ðàçðåæåííîñòè, íàïðèìåð, ðàñïðåäåëåíèåÄèðèõëå [19,21]. Îòìåòèì, îäíàêî, ÷òî èñïîëüçîâàíèå ðàñïðåäåëåíèÿ Äèðèõëå ìîæåòñäåëàòü îïòèìèçàöèîííóþ çàäà÷ó íåâûïóêëîé, ÷òî çàòðóäíÿåò ïîèñê åå ðåøåíèÿ [20].6Âûâîäû íàñòîÿùåé äèïëîìíîé ðàáîòå áûëè èññëåäîâàíû è àïðîáèðîâàíû ìåòîäû óäàëåíèÿøóìà èç èçîáðàæåíèÿ ñ èñïîëüçîâàíèåì ðàçðåæåííîãî ïðåäñòàâëåíèÿ. Áûëà ïîêàçàíà ýâîëþöèÿ äàííûõ ìåòîäîâ îò ãëîáàëüíîãî ïîðîãîâîãî àëãîðèòìà (4) äî ñîâðåìåííûõ ìåòîäîâ, îñíîâàííûõ íà îáó÷åíèè ñëîâàðÿ (4.2) è êëàñòåðèçàöèè ïîõîæèõïàò÷åé èçîáðàæåíèÿ (17). Ìåòîäû áûëè àïðîáèðîâàíû íà èçîáðàæåíèÿõ ñ äîáàâëåííûì ñèíòåòè÷åñêè øóìîì; ñðàâíèâàëîñü èõ êà÷åñòâî.
Òàêæå, áûëà ïðîâåäåíà ñåðèÿýêñïåðèìåíòîâ íà ðåàëüíûõ èçîáðàæåíèÿõ, ïîëó÷åííûõ ñ òðàíñìèññèîííîãî ýëåêòðîííîãî ìèêðîñêîïà, è íà îáû÷íûõ íå÷åòêèõ ôîòîãðàôèÿõ. Ïðèâåäåííûå èññëåäîâàíèÿ ÿâëÿþòñÿ äîâîëüíî ïîëíûìè â äàííîé òåìàòèêå, è ìîãóò áûòü èñïîëüçîâàíûäëÿ äàëüíåéøåãî óëó÷øåíèÿ êà÷åñòâà óäàëåíèÿ øóìà ñ èñïîëüçîâàíèåì ðàçðåæåííîãî ïðåäñòàâëåíèÿ.Ñòàíäàðòíî, ïîèñê ðàçðåæåííîãî ïðåäñòàâëåíèÿ îñíîâûâàåòñÿ íà ðåøåíèè çàäà÷è îïòèìèçàöèè ñ ýìïèðè÷åñêèìè, ëèáî ýâðèñòè÷åñêèìè îãðàíè÷åíèÿìè. Ïîñòàíîâêà18(à) Èçîáðàæåíèå ñ óäàëåííûì øóìîì, KSVD(á) Èçîáðàæåíèå ñ óäàëåííûì øóìîì, CSR(â) Èçîáðàæåíèå ñ óäàëåííûì øóìîì, ïðåäëîæåííûé ìåòîäÐèñ. 23: Ñðàâíåíèå ïðåäëîæåííîãî ìåòîäà ñ ìåòîäàìè K-SVD è CSR íà èçîáðàæåíèèñσ = 20.19çàäà÷è ðàçëè÷àåòñÿ îò ìåòîäà ê ìåòîäó ïî èñïîëüçóåìûì ïðè îïòèìèçàöèè îãðàíè÷åíèÿì, è ïîýòîìó îíà îñíîâàíà íà íàáëþäàåìûõ ñâîéñòâàõ èçîáðàæåíèé.
Òàêîé ïîäõîä ÿâëÿåòñÿ åñòåñòâåííûì, îäíàêî îí ïëîõî îáîñíîâàí è áûëî íåîáõîäèìî âûÿñíèòüïðåäïîñûëêè ê èñïîëüçîâàíèþ ýòèõ ýìïèðè÷åñêèõ îãðàíè÷åíèé îïòèìèçàöèé. ìàãèñòåðñêîé ðàáîòå áûëà ïîñòðîåíà áàéåñîâñêàÿ ñõåìà, êîòîðàÿ ïðèâîäèò êôîðìóëèðîâêå çàäà÷ îïòèìèçàöèè ÷åðåç ìàêñèìóì àïîñòåðèîðíîé âåðîÿòíîñòè. Óêàçàííàÿ ñõåìà ïîçâîëÿåò äàòü èíòåðïðåòàöèþ èçâåñòíûõ ìåòîäîâ ñ òî÷êè çðåíèÿ ñòàòèñòè÷åñêèõ ìîäåëåé è ïðåäëàãàòü íîâûå ôîðìóëèðîâêè îïòèìèçàöèîííûõ çàäà÷,îñíîâûâàÿñü íà ÿñíûõ ìàòåìàòè÷åñêèõ ïðåäïîëîæåíèÿõ î ðàñïðåäåëåíèÿõ ñëó÷àéíûõ âåëè÷èí. Îïèñàííàÿ ñõåìà áûëà ïðèìåíåíà äëÿ íàèáîëåå ïðîäâèíóòîãî ìåòîäà,ïîçâîëèëà îáîñíîâàòü íåêîòîðûå åãî íåòî÷íîñòè, è äàæå ïðåäëîæèòü íîâóþ ôîðìóëèðîâêó çàäà÷è îïòèìèçàöèè äëÿ ýòîãî ìåòîäà.Ïðåäëîæåííàÿ ñõåìà ìîæåò áûòü èñïîëüçîâàíà â äàëüíåéøåé ðàáîòå ñ ðàçðåæåííûìè ïðåäñòàâëåíèÿìè, è ìîæåò áûòü ïðèìåíåíà íå òîëüêî äëÿ óäàëåíèÿ øóìà, íîè â äðóãèõ îáëàñòÿõ, ñâÿçàííûõ ñ àíàëèçîì èçîáðàæåíèé: âîññòàíîâëåíèå ïèêñåëåé,ïîâûøåíèå ðàçðåøåíèÿ è ñêàëèðîâàíèå.20Ñïèñîê ëèòåðàòóðû[1] Êîëäàåâà Ì. Óäàëåíèå øóìà èç èçîáðàæåíèÿ ñ èñïîëüçîâàíèåì ðàçðåæåííîãîïðåäñòàâëåíèÿ.
//Ñáîðíèê òåçèñîâ 56-é íàó÷íîé êîíôåðåíöèè ÌÔÒÈ, ñåêöèÿèíôîðìàòèêà - 2013.[2] Êîëäàåâà Ì. Óäàëåíèå øóìà èç èçîáðàæåíèÿ ñ èñïîëüçîâàíèåì ðàçðåæåííîãîïðåäñòàâëåíèÿ. Ìåòîäû, îñíîâàííûå íà îáó÷åíèè ñëîâàðÿ. //Ñáîðíèê òåçèñîâ57-é íàó÷íîé êîíôåðåíöèè ÌÔÒÈ, ñåêöèÿ èíôîðìàòèêà - 2014.[3] Elad M. Sparse and redundant representations: from theory to applications in signaland image processing. Springer, 2010.[4] Starck J.
L., Candes E. J., Donoho D. L. The curvelet transform for image denoising//Image Processing, IEEE Transactions on. 2002. Ò. 11. . 6. Ñ. 670-684.[5] Aharon M., Elad M. Image denoising via sparse and redundant representations overlearned dictionaries //Image Processing, IEEE Transactions on. 2006. Ò.
15. . 12. Ñ. 3736-3745[6] Chang S. G., Yu B., Vetterli M. Adaptive wavelet thresholding for image denoisingand compression //Image Processing, IEEE Transactions on. 2000. Ò. 9. . 9. Ñ. 1532-1546.[7] Guleryuz O. G. Nonlinear approximation based image recovery using adaptive sparsereconstructions and iterated denoising-part I: theory //Image Processing, IEEETransactions on. 2006. Ò. 15. . 3. Ñ. 539-554.[8] Hel-Or Y., Shaked D.
A discriminative approach for wavelet denoising //ImageProcessing, IEEE Transactions on. 2008. Ò. 17. . 4. Ñ. 443-457.[9] Mallat S. G., Zhang Z. Matching pursuits with time-frequency dictionaries //SignalProcessing, IEEE Transactions on. 1993. Ò. 41. .
12. Ñ. 3397-3415.[10] Chen S., Billings S. A., Luo W. Orthogonal least squares methods and theirapplication to non-linear system identication //International Journal of control. 1989. Ò. 50. . 5. Ñ. 1873-1896.[11] Gorodnitsky I. F., Rao B. D. Sparse signal reconstruction from limited data usingFOCUSS: A re-weighted minimum norm algorithm //Signal Processing, IEEETransactions on. 1997. Ò. 45. . 3. Ñ. 600-616.[12] Davis G., Mallat S., Avellaneda M. Adaptive greedy approximations //Constructiveapproximation. 1997. Ò. 13. . 1. Ñ.
57-98.[13] AharonM.,EladM.,BrucksteinA.K-SVD:AnAlgorithmforDesigningOvercomplete Dictionaries for Sparse Representation //Signal Processing, IEEETransactions on. 2006. Ò. 54. . 11. Ñ. 4311-4322.[14] Gersho A., Gray R. M. Vector quantization and signal compression. SpringerScience and Business Media, 1992.21[15] Dong W. et al. Sparsity-based image denoising via dictionary learning and structuralclustering//ComputerVisionandPatternRecognition(CVPR),2011IEEEConference on. IEEE, 2011.
Ñ. 457-464.[16] Candes E. J., Romberg J., Tao T. Robust uncertainty principles: Exact signalreconstruction from highly incomplete frequency information // Information Theory,IEEE Transactions on. 2006. Ò. 52. . 2. Ñ. 489-509.[17] Daubechies I., Defrise M., De Mol C. An iterative thresholding algorithm for linearinverse problems with a sparsity constraint // Communications on pure and appliedmathematics. 2004. Ò. 57. .
11. Ñ. 1413-1457.[18] Williams P. M. Bayesian regularization and pruning using a Laplace prior // Neuralcomputation. 1995. Ò. 7. . 1. Ñ. 117-143.[19] Bishop C. M. et al. Pattern recognition and machine learning. New York : Springer,2006. Vol. 4. No. 4. P. 12.[20] Boyd S., Vandenberghe L. Convex optimization. Cambridge university press, 2004.[21] Marlin B. M., Murphy K. P. Sparse Gaussian graphical models with unknown blockstructure // Proceedings of the 26th Annual International Conference on MachineLearning. ACM, 2009. Ñ. 705-712.22.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
