En filterbank består av flera filter som arbetar tillsammans för att dela upp en signal i olika frekvensband. De kan också användas för att slå ihop signaler separerade i frekvensplanet till en enda. Sedan tidigt 70-tal har man lärt sig att designa förlustfria filterbankar som alltså inte introducerar några som helst fel i systemet. Sådana filterbankar kallas PR-filterbankar, där PR står för 'perfekt rekonstruktion'. Exempel på applikationer där filterbankar används är bildkodning, audiokodning, kommunikationssystem och omvandling av analoga signaler till digitala (A/D-omvandling). Under de senaste åren har det framkommit att genom att lätta på kraven gällande perfekt rekonstruktion, går det att markant minska den erforderliga aritmetiska komplexiteten. Eftersom de flesta system i sig inte är förlustfria, kan man utan att egentligen påverka den totala prestandan tillåta små fel i filterbanken, så l¨ange dessa fel är försumbara i jämförelse med andra felkällor som t.ex. kvantisering och avrundning. Avhandlingen behandlar digitala filter och likformiga icke-PR-filterbankar. Merparten av filterbankarna är realiserade med någon slags moduleringsteknik (cosinus-, sinus- eller komplexmodulering). Den röda tråden genom avhandlingen är kombinationen av tämligen smala övergångsband och samtidigt låg aritmetisk komplexitet. Ett sätt att uppnå denna kombination är att använda sig av en teknik som heter frekvenssvarsmaskning och förkortas FRM. Denna metod har på ett framgångsrikt sätt använts i avhandlingen. En potentiell nackdel med FRMmetoden är att den medför en längre fördröjning genom systemet. Därför föreslås också ett sätt att syntetisera FRM-filter med låg fördröjning. Här optimeras filtren både med avseende på komplexitet och fördröjning samtidigt. En annan metod som utnyttjats för att kombinera relativt smala övergångsband med låg aritmetisk komplexitet är att använda IIR filter istället för FIR filter. Ett flertal exempel på filter och filterbankar, optimerade och syntetiserade i Matlab, illustrerar fördelarna med de föreslagna filter- och filterbanks-klasserna. / Filter banks are systems of several filters with a common input or a common output. They are used whenever a signal needs to be split into different frequency bands. Since the early seventies, the theory of digital filter banks has developed to a mature state. Today there exist numerous ways to design filter banks for different applications, such as image and audio coding, transmultiplexing in communication systems, echo cancellation, and analog-to-digital (A/D) conversion systems. However, earlier work has to a large extent been on the transfer function level, whereas in this thesis work, efficient realizations, important in e.g. low-power applications, are in focus. Further, most of the previous work have been focused on the perfect reconstruction (PR) case, which is, for many applications an unnecessarily severe restriction. It has been show that by relaxing the requirements on perfect reconstruction, and allowing the filter banks to have some errors, the arithmetic complexity can be reduced significantly. This thesis treats digital filters and uniform non-PR filter banks. A major part of the filter banks are realized using different modulation schemes (complex, cosine, or sine modulation). The governing idea through the thesis is the combination of frequency selectivity and low arithmetic complexity. One example on how to achieve frequency selective digital filters and filter banks with low arithmetic complexity is to use the frequency-response masking (FRM) approach. This approach together with the idea of using IIR filters instead of FIR filters is successfully used in the thesis. The price to pay for the reduced arithmetic complexity using FRM filters is unfortunately a longer overall delay. Therefore, some work has ben done in the field of low-delay FRM FIR filters as well. These filters are optimized on both low delay and low arithmetic complexity simultaneously. A number of design examples are included in order to demonstrate the benefits of the new classes of filters and filter banks.

Purpose: Previous studies have shown that some digital filters can enhance picture-image visibility for people with visual impairment. The ultimate purposes of this study are to determine the improvement of picture-image visibility for people with maculopathy using digital image enhancement, and to compare the enhancement effects of generic filters and custom-devised filters. The secondary interests are to investigate the effect of age and maculopathy on supra-threshold contrast matching and to investigate the spatial frequency characteristics of picture-images. Methods: In order to develop effective custom-devised filters, supra-threshold contrast matching and contrast thresholds for two age groups of subjects with normal vision (14 aged 20-50 years and 15 aged 51+ years) and three groups of people with maculopathy (13 with atrophic ARMD, 14 with exudative ARMD, and 8 with JMD) were measured. Amplitude spectrum at each spatial frequency and the slope of amplitude versus spatial frequency were measured to investigate the spatial frequency characteristics of single face and general scene images. To investigate the preference for filters, 7 generic filters and 4 custom-devised filters were applied to single faces and general scenes. The generic filters were high-pass/unsharp masking, contrast enhancement, Sobel edge enhancement, DoG convolution, DoG FFT, Peli’s adaptive enhancement, and a band-pass filter with equi-emphasis of spatial frequencies. The custom-devised filters were band-pass filters based on contrast sensitivity (CS) loss, contrast matching at 3.6% and 27.9%, and emphasis of the peak of the CS curve. Subjects with maculopathy were required to rate the visibility of each image with and without filtering. Nine subjects with maculopathy participated to assess the enhancement quantitatively during which the recognition of facial expression and details in general scenes was tested with and without filtering. Results: Contrast constancy was demonstrated in age-matched controls and people with maculopathy. Single faces were found to be of significantly lower average amplitude than the other groups of images. Eight filters were found to be effective in improving perceived visibility; contrast enhancement, Peli’s adaptive enhancement, DoG convolution, high-pass/unsharp masking, Sobel edge enhancement, band-pass based on 3.6% and 27.9% contrast matching and equi-emphasis band-pass filters. These filters specifically were found to be effective for one or more combinations of maculopathy type and image category. The most commonly preferred filters were the generic filters, contrast enhancement and Peli’s adaptive enhancement. The two highest rated filters for each subject significantly reduced the number of errors of facial expression and errors of recognition of detail within general scene images. Conclusions: The visual system adjusts to compensate for CS loss with aging and maculopathy. Single faces are unique in spatial frequency characteristics. Some generic and custom-devised filters are effective in enhancing image visibility. The custom-devised filters are not superior to the generic filters. Visibility enhancement can be assessed quantitatively.

The goal of this bachelor thesis has been to investigate if the laboratory exercises in the courses digital filters and linear systems can be moved from matlab to a digital signal processor. The processor is a TMS320C6713 floating point processor mounted on a development board.   The original laboratories have been implemented and analyzed and some suggested changes have been presented for the digital filter laboration. For the laboration in linear systems, the exercise can be implemented as it is today. Furthermore, a transmultiplexer has been implemented and tested for real time execution.   Finally, an application programming interface has also been implemented, with common functions, used in the laboratories.

Purpose: Previous studies have shown that some digital filters can enhance picture-image visibility for people with visual impairment. The ultimate purposes of this study are to determine the improvement of picture-image visibility for people with maculopathy using digital image enhancement, and to compare the enhancement effects of generic filters and custom-devised filters. The secondary interests are to investigate the effect of age and maculopathy on supra-threshold contrast matching and to investigate the spatial frequency characteristics of picture-images. Methods: In order to develop effective custom-devised filters, supra-threshold contrast matching and contrast thresholds for two age groups of subjects with normal vision (14 aged 20-50 years and 15 aged 51+ years) and three groups of people with maculopathy (13 with atrophic ARMD, 14 with exudative ARMD, and 8 with JMD) were measured. Amplitude spectrum at each spatial frequency and the slope of amplitude versus spatial frequency were measured to investigate the spatial frequency characteristics of single face and general scene images. To investigate the preference for filters, 7 generic filters and 4 custom-devised filters were applied to single faces and general scenes. The generic filters were high-pass/unsharp masking, contrast enhancement, Sobel edge enhancement, DoG convolution, DoG FFT, Peli’s adaptive enhancement, and a band-pass filter with equi-emphasis of spatial frequencies. The custom-devised filters were band-pass filters based on contrast sensitivity (CS) loss, contrast matching at 3.6% and 27.9%, and emphasis of the peak of the CS curve. Subjects with maculopathy were required to rate the visibility of each image with and without filtering. Nine subjects with maculopathy participated to assess the enhancement quantitatively during which the recognition of facial expression and details in general scenes was tested with and without filtering. Results: Contrast constancy was demonstrated in age-matched controls and people with maculopathy. Single faces were found to be of significantly lower average amplitude than the other groups of images. Eight filters were found to be effective in improving perceived visibility; contrast enhancement, Peli’s adaptive enhancement, DoG convolution, high-pass/unsharp masking, Sobel edge enhancement, band-pass based on 3.6% and 27.9% contrast matching and equi-emphasis band-pass filters. These filters specifically were found to be effective for one or more combinations of maculopathy type and image category. The most commonly preferred filters were the generic filters, contrast enhancement and Peli’s adaptive enhancement. The two highest rated filters for each subject significantly reduced the number of errors of facial expression and errors of recognition of detail within general scene images. Conclusions: The visual system adjusts to compensate for CS loss with aging and maculopathy. Single faces are unique in spatial frequency characteristics. Some generic and custom-devised filters are effective in enhancing image visibility. The custom-devised filters are not superior to the generic filters. Visibility enhancement can be assessed quantitatively.

Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2004. / Includes bibliographical references (leaves 108-111). Also available in electronic version. Access restricted to campus users.

Thesis (MSc)--University of Stellenbosch, 2008. / Bibliography. Also available via the Internet.

published_or_final_version / abstract / toc / Electrical and Electronic Engineering / Master / Master of Philosophy

published_or_final_version / Electrical and Electronic Engineering / Master / Master of Philosophy

No description available.

No description available.