{"id":965,"date":"2023-03-14T16:16:25","date_gmt":"2023-03-14T13:16:25","guid":{"rendered":"https:\/\/legendhomework.com\/tr\/?p=965"},"modified":"2023-03-14T16:16:25","modified_gmt":"2023-03-14T13:16:25","slug":"goruntuleri-cozumleme-multimedya-bolumu-multimedya-bolumu-odevleri-multimedya-bolumu-tez-yaptirma-multimedya-bolumu-odev-ucretleri","status":"publish","type":"post","link":"https:\/\/legendhomework.com\/tr\/goruntuleri-cozumleme-multimedya-bolumu-multimedya-bolumu-odevleri-multimedya-bolumu-tez-yaptirma-multimedya-bolumu-odev-ucretleri\/","title":{"rendered":"G\u00f6r\u00fcnt\u00fcleri \u00c7\u00f6z\u00fcmleme \u2013 Multimedya B\u00f6l\u00fcm\u00fc \u2013 Multimedya B\u00f6l\u00fcm\u00fc \u00d6devleri \u2013 Multimedya B\u00f6l\u00fcm\u00fc Tez Yapt\u0131rma \u2013Multimedya B\u00f6l\u00fcm\u00fc \u00d6dev \u00dccretleri"},"content":{"rendered":"

G\u00f6r\u00fcnt\u00fcleri \u00c7\u00f6z\u00fcmleme<\/span><\/h3>\n

G\u00f6sterilen y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc bir g\u00f6r\u00fcnt\u00fcn\u00fcn olu\u015fturulmas\u0131, e\u011fitim a\u015famas\u0131nda elde edilen parametrelerle d\u00fc\u015f\u00fck \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc giri\u015f g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fcn kom\u015fuluk yap\u0131s\u0131n\u0131n d\u00f6n\u00fc\u015ft\u00fcr\u00fclmesinden kaynaklan\u0131r.<\/span><\/p>\n

Kom\u015fulu\u011fun ortalamas\u0131 daha sonra d\u00f6n\u00fc\u015f\u00fcme geri eklenir. LAM’ler kullan\u0131ld\u0131\u011f\u0131nda, d\u00fc\u015f\u00fck \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc bir kom\u015fu xr’nin s\u00fcper \u00e7\u00f6z\u00fcn\u00fcrl\u00fc\u011f\u00fc, burada Wc ve bc’nin cth LAM ile ili\u015fkili a\u011f\u0131rl\u0131k matrisi ve sapma vekt\u00f6r\u00fc oldu\u011fu \u015fekilde ifade edilebilir.<\/span><\/p>\n

Daha \u00f6nce tart\u0131\u015f\u0131ld\u0131\u011f\u0131 gibi, denklem (4.19) ile denklem (4.7) aras\u0131nda do\u011frudan bir ili\u015fki vard\u0131r. Denklem (4.19), y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc kom\u015fuluklar\u0131n yap\u0131s\u0131n\u0131 s\u02c6r olu\u015fturur. Alt simge r, in\u015fa edilmekte olan mahalleyi ifade eder. \u00c7ak\u0131\u015fan yap\u0131land\u0131r\u0131lm\u0131\u015f piksellerin ortalamas\u0131 al\u0131n\u0131r ve y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc g\u00f6r\u00fcnt\u00fc bu \u015fekilde olu\u015fturulur. Birka\u00e7 y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc \u00f6rne\u011fin ortalamas\u0131n\u0131n al\u0131nmas\u0131, nihai y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc \u00f6rne\u011fin g\u00fcvenilirli\u011fini art\u0131r\u0131r.<\/span><\/p>\n

Denklem (4.19), L = (2G1 – 1)(2G2 – 1); xe, geni\u015fletilmi\u015f d\u00fc\u015f\u00fck \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc g\u00f6r\u00fcnt\u00fcd\u00fcr.<\/span>\u00a0A d\u00fc\u015f\u00fck \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc bir kom\u015fulu\u011fun ortalamas\u0131n\u0131 alan kc,l ile ayn\u0131 \u00f6l\u00e7\u00fcde sabit bir \u00e7ekirdektir (d\u00fcrt\u00fc tepki \u00f6rnekleri 1’e e\u015fittir); ve b[n ,n ] = b [n ]b [n ], birden fazla H1H2 12 1122 x\u02c6h[n1,n2]= s\u00fcper \u00e7\u00f6z\u00fcmlenmi\u015f numune tahmininin ortalamas\u0131n\u0131n al\u0131nmas\u0131ndan sorumludur.<\/span><\/p>\n

Bu b\u00f6l\u00fcmde g\u00f6sterilen sonu\u00e7lar, e\u011fitim i\u00e7in Peppers g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fc ve test i\u00e7in Lena g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fc kullan\u0131r. Lena g\u00f6r\u00fcnt\u00fcs\u00fc zaten resmedildi; Peppers g\u00f6r\u00fcnt\u00fcs\u00fc birka\u00e7 referansta bulunabilir.<\/span><\/p>\n

LAM tabanl\u0131 sonu\u00e7lar, bir kenar\u0131n var oldu\u011fu kabul edilen ve aksi takdirde \u00e7ift do\u011frusal bir enterpolasyon kullanan g\u00f6r\u00fcnt\u00fc b\u00f6lgeleri boyunca ideal bir ad\u0131m kenar\u0131na uyan alt piksel kenar lokalizasyonu ve enterpolasyon (SEL) tekni\u011fi dahil olmak \u00fczere \u00e7e\u015fitli \u00e7ekirdek tabanl\u0131 enterpolasyon sonu\u00e7lar\u0131yla kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131. SEL tekni\u011fi i\u00e7in parametreler bildirilenlerle ayn\u0131yd\u0131.<\/span><\/p>\n

Lena ve Peppers 128 \u00d7 128 g\u00f6r\u00fcnt\u00fclerinin \u00e7ekirdek tabanl\u0131 enterpolasyonundan kaynaklanan en y\u00fcksek sinyal-g\u00fcr\u00fclt\u00fc oran\u0131 (PSNR) hakk\u0131nda raporlar, her boyutta 2 kat.<\/span><\/p>\n

\u00c7izim, Lena 128 \u00d7 128 g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fc her boyutta 2 kat de\u011fi\u015fen LAM say\u0131lar\u0131 ile \u00fcst \u00e7\u00f6z\u00fcmleme yaparken PSNR’yi g\u00f6sterir. Sistem parametreleri (\u00f6zellik vekt\u00f6rleri, a\u011f\u0131rl\u0131klar ve \u00f6nyarg\u0131lar), Peppers 256 \u00d7 256 g\u00f6r\u00fcnt\u00fcs\u00fc (yani, farkl\u0131 bir g\u00f6r\u00fcnt\u00fc) kullan\u0131larak e\u011fitildi.<\/span><\/p>\n

Kesintisiz ve kesikli \u00e7izgiler, destek b\u00f6lgeleri (ROS) 3 \u00d7 3 ve 5 \u00d7 5 kullan\u0131larak s\u0131ras\u0131yla e\u011fitim ve test seti yeniden yap\u0131land\u0131rma performans\u0131n\u0131 g\u00f6sterir. Genel olarak, e\u011fitim setinin PSNR’si, LAM say\u0131s\u0131 artt\u0131k\u00e7a artt\u0131. Bu, sezgisel olarak beklenen bir durumdur \u00e7\u00fcnk\u00fc LAM’lerin say\u0131s\u0131ndaki bir art\u0131\u015f, belirli g\u00f6r\u00fcnt\u00fc \u00f6zelliklerinde daha fazla uzmanla\u015fma sa\u011flar, dolay\u0131s\u0131yla daha do\u011fru bir g\u00f6r\u00fcnt\u00fc rekonstr\u00fcksiyonu sa\u011flar.<\/span><\/p>\n

\n

G\u00fcvenlik kamera g\u00f6r\u00fcnt\u00fcs\u00fc NETLE\u015eT\u0130RME<\/span>
\nKriminal G\u00f6r\u00fcnt\u00fc netle\u015ftirme<\/span>
\nVideo \u00e7\u00f6z\u00fcmleme<\/a> tutana\u011f\u0131<\/span>
\nPolislerin kulland\u0131\u011f\u0131 g\u00f6r\u00fcnt\u00fc netle\u015ftirme<\/span>
\nCd \u00e7\u00f6z\u00fcmleme bilirki\u015fi<\/span>
\nG\u00fcvenlik kameras\u0131 g\u00f6r\u00fcnt\u00fc kalitesi artt\u0131rma<\/span>
\nCD \u00e7\u00f6z\u00fcmleme tutana\u011f\u0131 \u00d6rne\u011fi<\/span>
\nKamera g\u00f6r\u00fcnt\u00fc NETLE\u015eT\u0130RME program\u0131<\/span><\/p>\n

\n

5 \u00d7 5 ROS kullan\u0131larak \u00e7\u0131kar\u0131lan \u00f6zellik seti, 3 \u00d7 3 ROS’tan daha fazla makroskopik g\u00f6r\u00fcnt\u00fc \u00f6zelli\u011fi sa\u011flar. Bu, ilgilenilen g\u00f6r\u00fcnt\u00fcye \u00f6zg\u00fc \u00f6zelliklerin daha fazla uzmanla\u015fmas\u0131na ve genellikle e\u011fitim setinde daha sad\u0131k bir g\u00f6r\u00fcnt\u00fc yeniden yap\u0131land\u0131rmas\u0131na neden olur.<\/span><\/p>\n

Bununla birlikte, test setinde, daha b\u00fcy\u00fck ROS, g\u00f6r\u00fcnt\u00fc \u00f6zelliklerinde uzmanla\u015fma derecesi artt\u0131k\u00e7a PSNR performans\u0131nda bir d\u00fc\u015f\u00fc\u015f g\u00f6sterme e\u011filimindeydi. Yapt\u0131\u011f\u0131m\u0131z t\u00fcm testlerde bu genel e\u011filimle kar\u015f\u0131la\u015f\u0131ld\u0131. Sistem daha fazla uzmanla\u015ft\u0131k\u00e7a (daha fazla LAM kullan\u0131r) \u00f6zellikler aras\u0131ndaki benzerli\u011fin daha mikroskobik d\u00fczeyde ortaya \u00e7\u0131kma e\u011filiminde oldu\u011funu \u00f6ne s\u00fcrer. \u00c7ekirdek ailesi yakla\u015f\u0131m\u0131n\u0131n, listelenen y\u00f6ntemlerden daha y\u00fcksek PSNR verdi\u011fi de g\u00f6zlemlenebilir.<\/span><\/p>\n

Lena g\u00f6r\u00fcnt\u00fcs\u00fc i\u00e7in ortak yakla\u015f\u0131mlar ve \u00e7ekirdek aile yakla\u015f\u0131m\u0131 kullan\u0131larak sonu\u00e7lar\u0131n g\u00f6rsel bir kar\u015f\u0131la\u015ft\u0131rmas\u0131 \u015eekil 4.9’da g\u00f6r\u00fclebilir. Bu \u015fekillerin her birinde g\u00f6sterilen e\u011fitim ve test g\u00f6r\u00fcnt\u00fcleri, 30 LAM ve 3 \u00d7 3 ROS kullan\u0131larak olu\u015fturulmu\u015ftur.<\/span><\/p>\n

Bir daire ile i\u015faretlenmi\u015f noktalara kar\u015f\u0131l\u0131k gelirler. Test g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fcn yeniden olu\u015fturulmas\u0131nda kullan\u0131lan Peppers 128 \u00d7 128 g\u00f6r\u00fcnt\u00fcs\u00fcnden \u00e7\u0131kar\u0131lan 30 \u00f6zelli\u011fi g\u00f6r\u00fcyoruz. Bu \u00f6zelliklerin ne kadar “d\u00fczenli” ve sinirli oldu\u011funa dikkat edin. \u00c7\u0131kar\u0131lan \u00f6znitelikler g\u00f6r\u00fcnt\u00fcye ba\u011fl\u0131d\u0131r ve \u00f6rne\u011fin, doku g\u00f6r\u00fcnt\u00fclerinden farkl\u0131 bir k\u00fcmenin sonu\u00e7lanmas\u0131n\u0131 bekleriz.<\/span><\/p>\n

Her bir \u00f6zelli\u011fin alt\u0131ndaki say\u0131, her bir \u00f6zellikte bulunan en b\u00fcy\u00fck ve en k\u00fc\u00e7\u00fck de\u011fer aras\u0131ndaki maksimum gri d\u00fczey fark\u0131n\u0131 g\u00f6sterir. Bu nedenle, alt\u0131nda \u201c1\u201d bulunan \u00f6zellik, sabit bir \u00f6zellik olarak kabul edilebilir, yani pratikte hi\u00e7bir yap\u0131 i\u00e7ermeyen bir \u00f6zelliktir.<\/span><\/p>\n

Bu \u00f6zellik, \u00e7ok d\u00fczg\u00fcn olan g\u00f6r\u00fcnt\u00fc b\u00f6l\u00fcmlerine kar\u015f\u0131l\u0131k gelir. \u00d6zellikler burada \u00f6l\u00e7eklenmemi\u015ftir; bunun yerine, sabit \u00f6zellik (s\u0131f\u0131r vekt\u00f6r\u00fc olan) gri ile temsil edilir (8 bitlik \u00f6l\u00e7ekte 128). Pozitif \u00f6zellik de\u011ferleri daha a\u00e7\u0131k hale gelir ve negatif \u00f6zellik de\u011ferleri orant\u0131l\u0131 olarak daha koyu bir gri tonuyla temsil edilir.<\/span><\/p>\n

\u015eekil 4.9’daki a\u015f\u0131r\u0131 \u00e7\u00f6z\u00fcmlenmi\u015f e\u011fitim ve test g\u00f6r\u00fcnt\u00fcleri benzer kalitedeydi. \u00c7ekirdek ailesi s\u00fcper-\u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc g\u00f6r\u00fcnt\u00fcler, burada sunulan di\u011fer yakla\u015f\u0131mlarla elde edilenlerden daha net g\u00f6r\u00fcn\u00fcr.<\/span><\/p>\n

Kar\u015f\u0131l\u0131k gelen yeniden olu\u015fturulmu\u015f g\u00f6r\u00fcnt\u00fclerin b\u00fcy\u00fckl\u00fck spektrumlar\u0131n\u0131 g\u00f6sterdik. Buradaki spektrumlar, yaln\u0131zca \u015fekilde g\u00f6sterilen k\u0131s\u0131m i\u00e7in de\u011fil, tamamen yeniden olu\u015fturulmu\u015f g\u00f6r\u00fcnt\u00fc i\u00e7indir. Bu spektrumlar\u0131n g\u00f6r\u00fcnt\u00fclenmesinden, LAM tabanl\u0131 yakla\u015f\u0131m\u0131n \u00f6rnekleme frekans\u0131n\u0131n yar\u0131s\u0131n\u0131n \u00fczerindeki bilgileri kurtard\u0131\u011f\u0131 ve orijinal g\u00f6r\u00fcnt\u00fclerin y\u00fcksek frekansl\u0131 bilgi \u00f6zelli\u011fini daha iyi yeniden \u00fcretti\u011fi a\u00e7\u0131kt\u0131r.<\/span><\/p>\n

SEL yakla\u015f\u0131m\u0131 ayr\u0131ca y\u00fcksek frekansl\u0131 bilgileri yeniden \u00fcretebilir. Bunun nedeni, daha \u00f6nce belirtildi\u011fi gibi, SEL yakla\u015f\u0131m\u0131n\u0131n kenar olarak kabul etti\u011fi g\u00f6r\u00fcnt\u00fc b\u00f6lgeleri boyunca ideal bir ad\u0131m kenar\u0131na uymas\u0131d\u0131r. SEL yakla\u015f\u0131m\u0131n\u0131n d\u00fc\u015f\u00fck PSNR’si, sorunsuz de\u011fi\u015fen g\u00f6r\u00fcnt\u00fc b\u00f6lgelerindeki performans\u0131na ba\u011flanabilir.<\/span><\/p>\n

Bunun nedeni, bu b\u00f6lgelerde SEL yakla\u015f\u0131m\u0131n\u0131n enterpolasyonu i\u00e7in \u00e7ift do\u011frusal \u00e7ekirde\u011fi kullanmas\u0131d\u0131r. \u00d6zetle, bu \u00e7al\u0131\u015fman\u0131n s\u00fcper-\u00e7\u00f6z\u00fclm\u00fc\u015f g\u00f6r\u00fcnt\u00fcleri genellikle burada sunulan di\u011fer yakla\u015f\u0131mlarla elde edilenlerden daha net g\u00f6r\u00fcn\u00fcr. Yakla\u015f\u0131m\u0131m\u0131zla kenarlar iyi korunmu\u015f gibi g\u00f6r\u00fcn\u00fcyordu ve metodolojimizle elde edilen daha y\u00fcksek PSNR, burada bildirilen di\u011fer yakla\u015f\u0131mlara g\u00f6re sorunsuz de\u011fi\u015fen g\u00f6r\u00fcnt\u00fc b\u00f6lgelerinde yeniden yap\u0131land\u0131rman\u0131n do\u011frulu\u011funun kan\u0131t\u0131d\u0131r.<\/span><\/p>\n

Sistemin \u00f6l\u00e7ekler aras\u0131nda fazlal\u0131\u011f\u0131 iyi bir \u015fekilde yakalad\u0131\u011f\u0131 hipotezimizi test etmek i\u00e7in, Lena128\u00d7128 g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fcn s\u00fcper \u00e7\u00f6z\u00fcn\u00fcrl\u00fc\u011f\u00fcnde, ayn\u0131 kod \u00e7izelgesi ve LAM’ler ile iki ard\u0131\u015f\u0131k G1 = G2 = 2 yeniden yap\u0131land\u0131rma a\u015famas\u0131n\u0131 kullanarak g\u00f6steriyoruz.<\/span><\/p>\n

Ba\u015fka bir deyi\u015fle, ortaya \u00e7\u0131kan “test” g\u00f6r\u00fcnt\u00fcs\u00fc, her boyutta toplam 4’l\u00fck bir s\u00fcper \u00e7\u00f6z\u00fcn\u00fcrl\u00fck fakt\u00f6r\u00fc i\u00e7in ilk s\u00fcper \u00e7\u00f6z\u00fcn\u00fcrl\u00fck a\u015famas\u0131nda kullan\u0131lan ayn\u0131 parametrelerle sistemden iki kez beslenir.<\/span><\/p>\n

LAM yeniden yap\u0131land\u0131r\u0131lm\u0131\u015f g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fcn, di\u011fer geni\u015fletilmi\u015f g\u00f6r\u00fcnt\u00fclerden daha net oldu\u011funa dikkat edin. Bu ayn\u0131 zamanda, s\u00fcper \u00e7\u00f6z\u00fcn\u00fcrl\u00fck i\u00e7in kulland\u0131\u011f\u0131m\u0131z \u00f6l\u00e7eklerdeki g\u00f6r\u00fcnt\u00fc kom\u015fuluklar\u0131n\u0131n benzerli\u011fine ili\u015fkin iddiam\u0131z\u0131 da desteklemektedir. G\u00f6r\u00fcnt\u00fc yeniden yap\u0131land\u0131rmas\u0131 i\u00e7in uygun olmayan baz setleri kullan\u0131ld\u0131\u011f\u0131nda neler olabilece\u011fine dair bir durumu g\u00f6sterir.<\/span><\/p>\n

Bu \u015fekilde, 128<\/span>\u00a0\u00d7 128 g\u00f6r\u00fcnt\u00fc 256 \u00d7 256 olarak yeniden olu\u015fturuluyor. \u0130stenilen g\u00f6r\u00fcnt\u00fc veriliyor. Peppers 128\u00d7128 g\u00f6r\u00fcnt\u00fcs\u00fcnden Peppers 256\u00d7256 g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fc yeniden olu\u015fturmak i\u00e7in e\u011fitimde elde edilen 30 \u00f6zellik ve LAM’den yeniden olu\u015fturulan g\u00f6r\u00fcnt\u00fc ve Pentagon 256\u00d7256’y\u0131 yeniden yap\u0131land\u0131rmak i\u00e7in e\u011fitimde elde edilen 30 \u00f6zellik ve LAM’den yeniden olu\u015fturulan g\u00f6r\u00fcnt\u00fc 128\u00d7128 g\u00f6r\u00fcnt\u00fc veriliyor.<\/span><\/p>\n

\n\n
<\/div>\n\n<\/div>\n \n