Title (srp)

Нови модел парцијалног филтрирања у реализацији алгоритама за детекцију ивица и сегментацију дигиталне слике

Author

Ивковић, М.Ратко

Contributor

Петровић, Миле

Description (srp)

Ova disertacija je doprinos digitalnoj analizi i obradi slike. Problematika koja je obrađena u disertaciji pokriva oblasti ocene kvaliteta, detekcije ivica, restauracije, klaster filtriranja, klasifikacije, superrezolucije, dizajna filtera i filtriranja digitalne slike. Za primenu u svim pomenutim oblastima razvijen je, a u disertaciji detaljno opisan novi metod parcijalnog filtriranja digitalne slike ‒ metod mozaika. Takođe, predstavljen je i model detekcije ivica ‒ hibridni metod ‒ koji čini sastavni deo metoda mozaika. Detaljno su analizirani parametri ocene kvaliteta. Na taj način rezultati disertacije predstavljeni su na adekvatan i sa drugim radovima merljiv način. Zbog preciznosti ocene filtriranja razvijen je model za ocenu sličnosti slike po kanalima – CSI. Dobijeni rezultati u disertaciji vrednovani su numerički na osnovu relevantnih parametara za ocenu kvaliteta multimedijalnih signala kao što su: PSNR, MSE, SNR, entropije, LoD, SSIM, MSSIM, DSSIM i CSI. Zasnovan na detaljnoj analizi algoritama detekcije ivica, kao još jedan doprinos disertacije, predložen je hibridni metod detekcije ivica. Upotrebom metoda mozaika izvršena je restauracija digitalne slike različitim klaster filtriranjem. Rezultati su prikazani nad slikama snimljenim niskim stepenom osvetljenja, kao i nad defokusiranim i zamućenim slikama. Adekvatnom analizom i obradom izvršena je klasifikacija segmenata u odnosu na parametar nivoa detalja. Praktična primena urađena je na BI-RADS medicinskim slikama. Superrezolucija digitalne slike izvršena je segmentacijom i klasifikacijom segmenata u okviru metoda mozaika. Analizom statističke vrednosti okoline piksela predložen je model za procenu koncentracije Snow & Rain šuma i dizajnirani su filteri za Snow & Rain i Salt & Papper šum. Modeli opisani u disertaciji testirani su korišćenjem poslednjih verzija softverskih rešenja kao što su Matlab, VCDemo, CVIPTools, Gimp, ImageQualityMeasurement, NeatImagePro i SofAS.

Description (eng)

This dissertation represents a contribution to digital image processing and analysis. The research issues in the dissertation are image quality estimation, edge detection, image restoration, cluster filtering, classification, super-resolution, filter design and filtering of digital image processing. For the application in all the mentioned fields, it was developed a new method of partial filtering of digital image - the mosaic method - and it was described in detail in the dissertation. Also, a model of edge detection was presented as the hybrid method, which is structural part of the mosaic method. Quality estimation parameters were analysed in details. In this way, results of the dissertation are presented in an appropriate, and with other works, measurable way. Due to its filtration assessment accuracy, there was developed model for assessment similarities of images on channels - CSI. Obtained results were numerically valuable based following on relevant parameters of signal such as: PSNR, MSE, SNR, Entropy, LoD, SSIM, MSSIM, DSSIM and CSI. Based on a detailed analysis of edge detection algorithms, a hybrid edge detection method is proposed as another contribution to the dissertation. Using a mosaic method, the restauration of the digital image was done by different cluster filtration. The results were shown on the images generated on the low level of brightness, as well as defocused and blurred images. By using an appropriate analysis and interpretation, it was done classification of segments based on the level of details as a parameter. Practical use is carried out on the BI-RADS medical images. Super resolution of the digital image was conducted by segmentation and by classifying segments in the mosaic method. Applying the statistical values of the pixels environment it was suggested the model for concentration assessment Snow & Rain noise and the filters for Snow & Rain and Salt & Pepper noise. These models, described in the dissertation, were tested by using the latest versions of softer technology such as Matlab, VC Demo, CVIPTools, Gimp, ImageQualityMeasurement, NeatImagePro i SofAS.

Object languages

Serbian

Date

2019

Rights

Creative Commons License
This work is licensed under a
CC BY-NC-SA 2.0 AT - Creative Commons Attribution - Non-Commercial - Share Alike 2.0 Austria License.

CC BY-NC-SA 2.0 AT

http://creativecommons.org/licenses/by-nc-sa/2.0/at/