Durağan görüntü sıkıştırma yöntemleri
Başlık çevirisi mevcut değil.
- Tez No: 46450
- Danışmanlar: DOÇ.DR. BÜLENT ÖRENCİK
- Tez Türü: Yüksek Lisans
- Konular: Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Computer Engineering and Computer Science and Control
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 1995
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Belirtilmemiş.
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 33
Özet
ÖZET Dijital görüntülerin kullanımı yaygınlaştıkça, saklama ve iletim için bilgiyi tamamen saklayan kayıpsız ve bir kısmını saklayan kayıplı yöntemler geliştirilmiştir. Bütün bu tekniklerin amacı görüntülerin fiziksel ortamda daha az yer kaplaması ve iletim sırasında daha az bilgi gönderilmesi yönündedir. Bu tezde her iki yöntemde incelenmiştir. Kayıpsız sıkıştırma yöntemi olarak entropi kodlaması üzerinde durulmuştur. Entropi kodlamasmda, sıkıştırma oranı, sıkıştırılacak görüntünün dağılımı ile doğru orantılıdır. Dağılımın kodlamaya daha uygun hale gelmesi için öngörü teknikleri önerilmiştir. Bu öngörü yöntemleri sonrasında oluşacak terimlerin kodlanması sonucu daha iyi sıkıştırma oranları elde edilebilir. İncelenen diğer bir teknik ise RLC yöntemidir. Koşu uzunluğunun çok olması durumunda uygulanabilecek bir yöntemdir. Bu yöntemin başarılı olması gene görüntü elemanlarına bağlıdır. Kayıplı sıkıştırma sonucunda oluşan görüntü or j inal kalitesinde olmasa da, görüntü kalitesi açısından orjinaline yakındır. Burada kullanılan yöntem, görüntünün bulunduğu domenden farklı bir domene aktarılıp orada kodlanmasıdır. Kullanılacak transformun sağlaması gereken en önemli özellik, görüntü enerjisini belirli bir bölgeye toplayabilmesidir. Enerjinin yoğun olduğu bölgede kodlama yapılarak yüksek oranda sıkıştırma sağlanabilir. Transform olarak DCT kullanılmıştır. Bloklara bölünen görüntü transform domenine aktarıldıktan sonra oluşan katsayılar kuantalanır ve Huffman Entropi Kodlaması yöntemiyle kodlanır. Sıkıştırma oranı seçilen blok boyuna ve kuantaya bağlıdır. iv
Özet (Çeviri)
SÜMMARY The increasing use of digital images has motivated studies in developing new technigues about storing and transmitting the image data. The main idea behind image compression is reducing the disk space allocation and reducing the information in transmission lines. There are two approaches in image compression. Öne is information preserving and the other öne is information reducing. This thesis includes both of them. Before defining the approaches, lef s explain the NMSE which gives the quality of an reconstructed image and entropy which is the minimum possible average bit rate required in coding a message. in fact entropy can be interpreted as the amount of information that a message contains. If xt is an image gray level and p() is the probability then entropy H (p) is 2fl-l H(p) =- J) p Uj) -Iog2p (xj If we assume that, L (p) bit/pixel coding is performed över a p distribution, then the measurement of our coding performance is given by the difference of E[L(p)] and H (p). The normalized version of image quality measurement MSE (Mean Sguare Error) is NMSE (Normalized Mean Sguare Error) is defined as ££ [x(i,j)-*(i,j)]2“*”' 'şş*».:»' vThe NMSE value being. zero means the reconstructed image and the original one are the same. Although MMSE gives us an idea about how good the reconstructed image is, it can't be a good measure. The main idea behind information preserving compression is, while using lower bit rates for the image, the reconstructed image has the same information as the original one. The thesis includes three types of information preserving compression techniques which are Huffman coding, predictive coding, run length coding. The first thing that must be done for compression is assigning some kind of code for every gray level in an image. Although assigning uniform length codes for gray levels is easy, smaller compression ratios are obtained by assigning variable length codes. While assigning the variable length codes, the most important thing is the designed codes must be uniquely decodable. That is they must be identified when they are received. In the Huffman coding few bits are assigned to frequently used elements, and more bits to less commonly used ones. In this case the average bit rate is reduced according to uniform length coding. As the coding involves with probabilities, the Huffman coding performance approaches to the value obtained by entropy. In predictive coding the distribution of the image data are changed. By this method, the newly created codes are distributed around zero. This decreases the value of entropy and as a results of this the ratio of compression increases. The estimated element is predicted from the previous elements of the image. The estimated prediction error is calculated as eyaxy-taai + ba2 + ca3) viwhere a,b,c are the previous elements and ax, a2, a3 are the coefficients. By using prediction error the amount of compression achieved increases. The collection of image data that have the same gray- value is called run. If there are long runs in the image then coding of the gray value and it's run length may lead us to compression. For example the image data values ( 100 100 100 100 100 50 50 50 ) maybe coded as (100,5) (50,3) pairs. The run length coding algorithm is suitable for binary images. In fact in binary images the only message coded will be the run lengths, not the image data values. The information reducing approach is the one whose encoding process will not keep all the original information, but retained image will have a high image quality. The approach used here is to transform the image into a other domain, different from the image intensity domain and perform the coding scheme there. The transform must have a high energy compaction property. The energy is packed into a certain zone in transform domain where the coding is performed. n^O n2=o «i-l jBT2-1 İCt-0 *2=0 The transform defined is a linear one. f (n^na) is the viiimage data, TfCk^kj) is the transform coefficients and a(.) b(.) are the basis functions than make the transform and inverse transform. In most cases these basis functions are separable that means the transform and inverse transforms may be computed by row-column decomposition. The row-column decomposition method reduces amount of mathematical computational needed. In transform coding the image is divided into subimages or blocks and every block is coded separately. The subimage coding reduces the computational requirements. As the sub image size decreases, the correlation among neighboring subimages increases. The transform coefficients are then quant it ized. As the quantization made on one coefficient in transform domain effects all the image data, the quantization process is very important. One of the most commonly used transform is DFT which has a good energy compaction property and a fast algorithm (FFT). It's possible to increase the energy compaction property of DFT without sacrificing other qualities. DCT has improved characteristics and is closely related to DFT. As DCT has less discontinuities then DFT, it has a better energy compaction property. The (0,0) element of the DCT coefficients carries most of the energy and named DC element. The others are AC elements and carry details of the subimage. The AC elements are Gaussian distributed that are suitable for Huffman coding. The importance of detail in subimage is in zigzag order. By changing the quantization matrix or by taking some of the AC elements in zigzag order the amount of compression changes. As the number of AC elements used reduces, the amount of detail decreases which causes subimage to blur and image to look blocky. viiiThe compressed data is stored on magnetic disk or transmitted to receiver. In both cases there is a possibility of a change in compressed data. Except RLC the compressed data is achieved by Huffman coding. As we know the Huffman coding and decoding depends on probability tree. If we assume that the Huffman tree is free of errors then change on a code leads us to a wrong data in decoding. Not only the current data but the next coming a number of data are also effected by an error. This also leads us to decode more or less number of image data. We can not correct this kind of an error completely. The error correction algorithm assumes that after a certain number of bits, a new image data is decoded. Even if an error occurs, because of the assumption above the image data decoded will be corrected. The error is kept into a region. IX
Benzer Tezler
- Görüntü sıkıştırma yöntemlerinin etkinliğini arttıran dönüşüm ve bölümlendirme işlemleri
Transformation and segmentation operations on images for increasing the effectiveness of compression methods
EMİR ÖZTÜRK
Yüksek Lisans
Türkçe
2012
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolTrakya ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
YRD. DOÇ. DR. ALTAN MESUT
- Video sıkıştırma yöntemleri üzerine karşılaştırmalı bir araştırma
A comparative study on video compression methods
VOLKAN CEYLAN
Yüksek Lisans
Türkçe
2004
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolEge ÜniversitesiUluslararası Bilgisayar Ana Bilim Dalı
PROF.DR. AYDIN ÖZTÜRK
- Darbantlı video iletişim sistemlerinde dalgacık dönüşümünün kullanılması
The use of wavelet transform in narrowband video communication systems
FARID RAJABLI
Yüksek Lisans
Türkçe
2005
Elektrik ve Elektronik MühendisliğiAnkara ÜniversitesiElektronik Mühendisliği Ana Bilim Dalı
PROF.DR. MÜMTAZ YILMAZ
- High impedance fault detection in medium voltage distribution systems using wavelet transform
Dalgacık dönüşümü kullanılarak orta gerilim dağıtım sistemlerinde yüksek empedanslı arıza tespiti
BARAA MAKKAWI
Yüksek Lisans
İngilizce
2024
Elektrik ve Elektronik Mühendisliğiİstanbul Teknik ÜniversitesiElektrik Mühendisliği Ana Bilim Dalı
PROF. DR. ÖMER USTA
- Video sinyallerinde hareket tespit uygulaması
Application of motion detection on video signals
KAAN GENÇAY
Yüksek Lisans
Türkçe
2019
Elektrik ve Elektronik MühendisliğiYıldız Teknik ÜniversitesiElektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
DOÇ. DR. UMUT ENGİN AYTEN