Spatial pattern analysis of thicket expansion in a semi-arid savanna
- Authors: Putzier, Rachel Rayne
- Date: 2024-10-11
- Subjects: Arid regions South Africa Eastern Cape , Spatial analysis (Statistics) , Thicket , Lidar , Cluster analysis , Trees Mortality , Scrub encroachment
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/464484 , vital:76515
- Description: Woody thickening has negative economic and ecological impacts in savannas globally. While the increase of savanna trees as a form of bush encroachment has been well studied, less is known about the increase of thicket species in savannas, which is an important phenomenon resulting in the formation of closed-canopy clumps which may exclude the grass layer over time. The early stage of thicket expansion is often initiated by a nucleating savanna species which facilitates the establishment of woody thicket pioneer species, and as the thicket clump formation expands, bush clumps increase in dominance, thereby increasing the frequency of competitive interactions and leading to a possible switch from facilitative to competitive interactions. Spatial point pattern analysis provides a useful tool to elucidate these underlying patterns and ecological processes. I used high resolution LiDAR data combined with spatial point pattern analysis to understand tree-tree interactions in a semi-arid savanna in the Eastern Cape Province of South Africa. I conducted a cluster analysis based on vegetation structural variables to distinguish different stages of woody plant encroachment from open savanna to closed canopy thicket. Using the canopy height model, I quantified the change in the grass height from open savanna to closed canopy thicket clumps as an indicator of a possible biome shift. Additionally, I used spatial point pattern analyses to investigate the effect of thicket clump formation on the composition of savanna and thicket species, the overall patterns of trees, and the associations of small thicket species with large Vachellia karroo trees, which serve as clump initiators. Finally, I examined the mortality of savanna trees across increasing stages of thicket expansion using second order spatial statistics, namely the Mark- and Pair-Correlation Functions. Results confirmed that three vegetation states, influenced by elevation, are present at the study site, representing open canopy savanna (early-stage thicket encroachment), encroached savanna with low thicket dominance (intermediate-stage thicket encroachment), and highly encroached with dominant thicket clumps (late-stage thicket encroachment). These stages showed increasing tree height, canopy cover and canopy height density, as well as decreased (but not completely absent grass layer) as thicket encroachment progresses. Spatial point pattern analysis showed, as predicted, that there was an overall aggregation of trees at small-scales within early thicket clump formation, from which I inferred that facilitative relationships may exist between trees. Contrary to my predictions, at later stages of thicket clump formation I found dominant independent patterns between savanna adults and juvenile thicket species, which may result from a combination of facilitative and competitive effects. Lastly, as expected, I found that the density of V. karroo mortality increased as thicket encroachment increased, with an overall random spatial pattern of dead V. karroo across encroachment stages. As predicted, tree mortality was randomly distributed in space in the open savanna state, and as thicket clump formation increases, tree competitive mortality became more evident, as well as decreased tree performance. Overall, the study highlights the interplay between facilitation and competition in semiarid savanna where thicket clumps are expanding. Intervention strategies are suggested to target areas of intermediate thicket clump formation, as these areas provide an opportunity to remove V. karroo before the nucleation process has enabled the establishment and increase of thicket species and to ensure the grass layer is kept productive. I conclude that the use of remote sensing and LiDAR technology holds a wide range of possibilities for monitoring and managing woody encroachment in savanna systems, however these methods need to be further refined for effective use within African savanna and thicket context, which displays high spatial aggregation making typical segmentation methods difficult. , Thesis (MSc) -- Faculty of Science, Botany, 2024
- Full Text:
- Date Issued: 2024-10-11
- Authors: Putzier, Rachel Rayne
- Date: 2024-10-11
- Subjects: Arid regions South Africa Eastern Cape , Spatial analysis (Statistics) , Thicket , Lidar , Cluster analysis , Trees Mortality , Scrub encroachment
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/464484 , vital:76515
- Description: Woody thickening has negative economic and ecological impacts in savannas globally. While the increase of savanna trees as a form of bush encroachment has been well studied, less is known about the increase of thicket species in savannas, which is an important phenomenon resulting in the formation of closed-canopy clumps which may exclude the grass layer over time. The early stage of thicket expansion is often initiated by a nucleating savanna species which facilitates the establishment of woody thicket pioneer species, and as the thicket clump formation expands, bush clumps increase in dominance, thereby increasing the frequency of competitive interactions and leading to a possible switch from facilitative to competitive interactions. Spatial point pattern analysis provides a useful tool to elucidate these underlying patterns and ecological processes. I used high resolution LiDAR data combined with spatial point pattern analysis to understand tree-tree interactions in a semi-arid savanna in the Eastern Cape Province of South Africa. I conducted a cluster analysis based on vegetation structural variables to distinguish different stages of woody plant encroachment from open savanna to closed canopy thicket. Using the canopy height model, I quantified the change in the grass height from open savanna to closed canopy thicket clumps as an indicator of a possible biome shift. Additionally, I used spatial point pattern analyses to investigate the effect of thicket clump formation on the composition of savanna and thicket species, the overall patterns of trees, and the associations of small thicket species with large Vachellia karroo trees, which serve as clump initiators. Finally, I examined the mortality of savanna trees across increasing stages of thicket expansion using second order spatial statistics, namely the Mark- and Pair-Correlation Functions. Results confirmed that three vegetation states, influenced by elevation, are present at the study site, representing open canopy savanna (early-stage thicket encroachment), encroached savanna with low thicket dominance (intermediate-stage thicket encroachment), and highly encroached with dominant thicket clumps (late-stage thicket encroachment). These stages showed increasing tree height, canopy cover and canopy height density, as well as decreased (but not completely absent grass layer) as thicket encroachment progresses. Spatial point pattern analysis showed, as predicted, that there was an overall aggregation of trees at small-scales within early thicket clump formation, from which I inferred that facilitative relationships may exist between trees. Contrary to my predictions, at later stages of thicket clump formation I found dominant independent patterns between savanna adults and juvenile thicket species, which may result from a combination of facilitative and competitive effects. Lastly, as expected, I found that the density of V. karroo mortality increased as thicket encroachment increased, with an overall random spatial pattern of dead V. karroo across encroachment stages. As predicted, tree mortality was randomly distributed in space in the open savanna state, and as thicket clump formation increases, tree competitive mortality became more evident, as well as decreased tree performance. Overall, the study highlights the interplay between facilitation and competition in semiarid savanna where thicket clumps are expanding. Intervention strategies are suggested to target areas of intermediate thicket clump formation, as these areas provide an opportunity to remove V. karroo before the nucleation process has enabled the establishment and increase of thicket species and to ensure the grass layer is kept productive. I conclude that the use of remote sensing and LiDAR technology holds a wide range of possibilities for monitoring and managing woody encroachment in savanna systems, however these methods need to be further refined for effective use within African savanna and thicket context, which displays high spatial aggregation making typical segmentation methods difficult. , Thesis (MSc) -- Faculty of Science, Botany, 2024
- Full Text:
- Date Issued: 2024-10-11
Cluster analysis for group selection in launch sales predictions
- Authors: Watchurst, Lee
- Date: 2021-04
- Subjects: Gqeberha (South Africa) , Eastern Cape (South Africa) , Cluster analysis
- Language: English
- Type: Master's theses , text
- Identifier: http://hdl.handle.net/10948/52003 , vital:43447
- Description: One way for businesses to stay ahead in a competitive market is through the launch of new products and planning for these launches optimally. This includes ordering the correct quantity of stock in advance as well as maintaining these stock levels while the item launches. However, holding too much stock in warehouses can affect the business costs adversely. This research proposes the use of cluster analysis techniques to determine the up-front purchase quantity by identifying similar items and using their initial quantities sold. Products will be grouped based on their numerical and categorical attributes. Once the data is clustered, the Bass model will be used to obtain a sales profile for the new item. The Bass model is a popular choice for product life cycle planning due to the emphasis placed on the timing of adoption. The study will make use of data from a retail and wholesale company that sells, in part, single use items. With the planning for new launches being a key problem point in many companies, this research aims to optimise the planning process and ensure product launch success across stores. , Thesis (MSc) -- Faculty of Science, School of Computer Science, Mathematics, Physics and Statistics, 2021
- Full Text: false
- Date Issued: 2021-04
- Authors: Watchurst, Lee
- Date: 2021-04
- Subjects: Gqeberha (South Africa) , Eastern Cape (South Africa) , Cluster analysis
- Language: English
- Type: Master's theses , text
- Identifier: http://hdl.handle.net/10948/52003 , vital:43447
- Description: One way for businesses to stay ahead in a competitive market is through the launch of new products and planning for these launches optimally. This includes ordering the correct quantity of stock in advance as well as maintaining these stock levels while the item launches. However, holding too much stock in warehouses can affect the business costs adversely. This research proposes the use of cluster analysis techniques to determine the up-front purchase quantity by identifying similar items and using their initial quantities sold. Products will be grouped based on their numerical and categorical attributes. Once the data is clustered, the Bass model will be used to obtain a sales profile for the new item. The Bass model is a popular choice for product life cycle planning due to the emphasis placed on the timing of adoption. The study will make use of data from a retail and wholesale company that sells, in part, single use items. With the planning for new launches being a key problem point in many companies, this research aims to optimise the planning process and ensure product launch success across stores. , Thesis (MSc) -- Faculty of Science, School of Computer Science, Mathematics, Physics and Statistics, 2021
- Full Text: false
- Date Issued: 2021-04
Parent characterization of quality protein maize (Zea mays L.) and combining ability for tolerance to drought stress
- Authors: Pfunde, Cleopatra Nyaradzo
- Date: 2012
- Subjects: Corn -- Quality , Corn as food , Corn -- Effect of stress on , Corn -- Effect of drought on , Cluster analysis , Crops -- Effect of drought on , Corn -- Drought tolerance , Corn -- Breeding , Crops -- Drought tolerance
- Language: English
- Type: Thesis , Masters , MSc Agric (Crop Science)
- Identifier: vital:11869 , http://hdl.handle.net/10353/d1007536 , Corn -- Quality , Corn as food , Corn -- Effect of stress on , Corn -- Effect of drought on , Cluster analysis , Crops -- Effect of drought on , Corn -- Drought tolerance , Corn -- Breeding , Crops -- Drought tolerance
- Description: Quality protein maize (QPM) has enhanced levels of two essential amino acids, lysine and tryptophan compared to normal maize. This makes QPM an important cereal crop in communities where maize is a staple crop. The main abiotic factor to QPM production is drought stress. Little information is available on the effect of drought stress on QPM. Therefore, the objectives of this study were to: (i) conduct diversity analysis of QPM inbred lines using morpho-agronomic and simple sequence repeat markers, (ii) screen available QPM inbred lines and F1 progeny for tolerance to seedling drought stress, (iii) determine the combining ability and type of gene action of QPM inbred lines for tolerance to seedling drought stress, grain yield and endosperm modification. The study was conducted in South Africa, at the University of Fort Hare. Morphological characterisation of 21 inbred lines was done using quantitative and qualitative traits. A randomised complete block design with three replicates was used for characterizing the inbred lines in the field. Genstat statistical software, version 12 (Genstat ®, 2009) was used for analysis of variance (ANOVA) and descriptive statistics. Analysis of variance was performed on all quantitative data for morphological traits. Data for qualitative traits was tabulated in their nominal classes. Traits that contributed most to the variation were days to anthesis, days to silking, anthesis-silking interval, plant height, number of kernel rows, ear length and grain yield. Cluster analysis grouped the inbred lines into three main clusters. The first cluster was characterised by tall and average yielding lines, while the second cluster showed the least anthesis-silking interval, and had the highest yield. Cluster three consisted of lines that were early maturing, but were the least yielding. Genetic distances between maize inbred lines were quantified by using 27 simple sequence repeat markers. The genetic distances between genotypes was computed using Roger’s (1972) genetic distances. Cluster analysis was then carried out using the neighbour-joining tree method using Power Marker software version 3.25. A dendrogram generated from the genetic study of the inbred lines revealed three groups that concurred with expectations based upon pedigree data. These groups were not identical to the groups generated using morpho-agronomic characterisation. Twenty one QPM inbred lines were crossed using a North Carolina design II mating scheme. These were divided into seven sets, each with three inbred lines. The three inbred lines in one set were used as females and crossed with three inbred lines in another set consisting of males. Each inbred line was used as a female in one set, and as a male in a second set. Sixty three hybrids (7 sets x 9 hybrids) were formed and evaluated in October 2011, using a 6x8 alpha-lattice incomplete block design with three replicates under glasshouse and optimum field conditions. A randomised complete block design with three replicates was used for the 21 parental inbred lines. Traits recorded for the glasshouse study were, canopy temperature, chlorophyll content, leaf roll, stem diameter, plant height, leaf number, leaf area, fresh and dry root and shoot weights. Data for the various traits for each environment, 25 percent (stress treatment) and 75 percent (non-stress) of field capacity, were subjected to analysis of variance using the unbalanced treatment design in Genstat statistical package Edition 12. Where varietal differences were found, means were separated using Tukey’s test. Genetic analyses for grain yield and agronomic traits were performed using a fixed effects model in JMP 10 following Residual Maximum Likelihood procedure (REML). From the results, inbred lines that were not previously classified into heterotic groups and drought tolerance categories were classified based on their total dry weight performance and drought susceptibility index. Inbred lines L18, L9, L8, L6 and L3, in order of their drought tolerance index were the best performers under greenhouse conditions and could be recommended for breeding new varieties that are tolerant to seedling drought stress. Evaluation of maize seedlings tolerant to drought stress under glasshouse conditions revealed that cross combination L18 x L11 was drought tolerant, while cross L20 x L7 was susceptible. Total dry weight was used as the major criteria for classifying F1 maize seedlings as being resistant or susceptible. General combining ability effects accounted for 67.43 percent of the genetic variation for total dry weight, while specific combining ability effects contributed 37.57 percent. This indicated that additive gene effects were more important than non-additive gene action in controlling this trait. In the field study (non-drought), the experimental design was a 6x8 alpha lattice incomplete block design with three replicates. On an adjacent field a randomised complete block design with three replicates was used to evaluate the parental inbred lines. The following variables were recorded: plant height, ear height, ears per plant, endosperm modification, days to silking and days to anthesis, anthesis-silking interval, number of kernels per row, number of rows per ear and grain yield. General analyses for the incomplete lattice block design and randomised complete block design for hybrid and inbred data respectively were performed using JMP 10 statistical software. Means were separated using the Tukey's test. Genetic analyses of data for grain yield and agronomic traits were conducted using a fixed effects model using REML in JMP 10. The importance of both GCA (51 percent) and SCA (49 percent) was observed for grain yield. A preponderance of GCA existed for ear height, days to anthesis, anthesis-silking interval, ears per plant and number of kernels per row, indicating that predominantly, additive gene effects controlled hybrid performance under optimum field conditions. The highest heritability was observed for days to silking (48.27 percent) suggesting that yield could be improved through selection for this trait. Under field conditions, variation in time to maturity was observed. This implies that these inbred lines can be recommended for utilisation in different agro-ecologies. Early maturing lines such as L18 can be used to introduce earliness in local cultivars, while early maturing single crosses such as L18 x L2, L5 x L9, L3 x L4 and L2 x L21 could be recommended for maize growers in drought prone areas such as the former Ciskei. Single crosses L18xL11, L16xL18, L8xL21 and L9xL6 had good tolerance to seedling drought stress. On the other hand, single crosses L18xL11 and L11xL13 had high grain yield and good endosperm modification. All these single crosses could be recommended for commercial production after evaluation across locations in the Eastern Cape Province. Alternatively they can be crossed with other superior inbreds to generate three or four way hybrids, which could then be evaluated for potential use by farmers in the Eastern Cape.
- Full Text:
- Date Issued: 2012
- Authors: Pfunde, Cleopatra Nyaradzo
- Date: 2012
- Subjects: Corn -- Quality , Corn as food , Corn -- Effect of stress on , Corn -- Effect of drought on , Cluster analysis , Crops -- Effect of drought on , Corn -- Drought tolerance , Corn -- Breeding , Crops -- Drought tolerance
- Language: English
- Type: Thesis , Masters , MSc Agric (Crop Science)
- Identifier: vital:11869 , http://hdl.handle.net/10353/d1007536 , Corn -- Quality , Corn as food , Corn -- Effect of stress on , Corn -- Effect of drought on , Cluster analysis , Crops -- Effect of drought on , Corn -- Drought tolerance , Corn -- Breeding , Crops -- Drought tolerance
- Description: Quality protein maize (QPM) has enhanced levels of two essential amino acids, lysine and tryptophan compared to normal maize. This makes QPM an important cereal crop in communities where maize is a staple crop. The main abiotic factor to QPM production is drought stress. Little information is available on the effect of drought stress on QPM. Therefore, the objectives of this study were to: (i) conduct diversity analysis of QPM inbred lines using morpho-agronomic and simple sequence repeat markers, (ii) screen available QPM inbred lines and F1 progeny for tolerance to seedling drought stress, (iii) determine the combining ability and type of gene action of QPM inbred lines for tolerance to seedling drought stress, grain yield and endosperm modification. The study was conducted in South Africa, at the University of Fort Hare. Morphological characterisation of 21 inbred lines was done using quantitative and qualitative traits. A randomised complete block design with three replicates was used for characterizing the inbred lines in the field. Genstat statistical software, version 12 (Genstat ®, 2009) was used for analysis of variance (ANOVA) and descriptive statistics. Analysis of variance was performed on all quantitative data for morphological traits. Data for qualitative traits was tabulated in their nominal classes. Traits that contributed most to the variation were days to anthesis, days to silking, anthesis-silking interval, plant height, number of kernel rows, ear length and grain yield. Cluster analysis grouped the inbred lines into three main clusters. The first cluster was characterised by tall and average yielding lines, while the second cluster showed the least anthesis-silking interval, and had the highest yield. Cluster three consisted of lines that were early maturing, but were the least yielding. Genetic distances between maize inbred lines were quantified by using 27 simple sequence repeat markers. The genetic distances between genotypes was computed using Roger’s (1972) genetic distances. Cluster analysis was then carried out using the neighbour-joining tree method using Power Marker software version 3.25. A dendrogram generated from the genetic study of the inbred lines revealed three groups that concurred with expectations based upon pedigree data. These groups were not identical to the groups generated using morpho-agronomic characterisation. Twenty one QPM inbred lines were crossed using a North Carolina design II mating scheme. These were divided into seven sets, each with three inbred lines. The three inbred lines in one set were used as females and crossed with three inbred lines in another set consisting of males. Each inbred line was used as a female in one set, and as a male in a second set. Sixty three hybrids (7 sets x 9 hybrids) were formed and evaluated in October 2011, using a 6x8 alpha-lattice incomplete block design with three replicates under glasshouse and optimum field conditions. A randomised complete block design with three replicates was used for the 21 parental inbred lines. Traits recorded for the glasshouse study were, canopy temperature, chlorophyll content, leaf roll, stem diameter, plant height, leaf number, leaf area, fresh and dry root and shoot weights. Data for the various traits for each environment, 25 percent (stress treatment) and 75 percent (non-stress) of field capacity, were subjected to analysis of variance using the unbalanced treatment design in Genstat statistical package Edition 12. Where varietal differences were found, means were separated using Tukey’s test. Genetic analyses for grain yield and agronomic traits were performed using a fixed effects model in JMP 10 following Residual Maximum Likelihood procedure (REML). From the results, inbred lines that were not previously classified into heterotic groups and drought tolerance categories were classified based on their total dry weight performance and drought susceptibility index. Inbred lines L18, L9, L8, L6 and L3, in order of their drought tolerance index were the best performers under greenhouse conditions and could be recommended for breeding new varieties that are tolerant to seedling drought stress. Evaluation of maize seedlings tolerant to drought stress under glasshouse conditions revealed that cross combination L18 x L11 was drought tolerant, while cross L20 x L7 was susceptible. Total dry weight was used as the major criteria for classifying F1 maize seedlings as being resistant or susceptible. General combining ability effects accounted for 67.43 percent of the genetic variation for total dry weight, while specific combining ability effects contributed 37.57 percent. This indicated that additive gene effects were more important than non-additive gene action in controlling this trait. In the field study (non-drought), the experimental design was a 6x8 alpha lattice incomplete block design with three replicates. On an adjacent field a randomised complete block design with three replicates was used to evaluate the parental inbred lines. The following variables were recorded: plant height, ear height, ears per plant, endosperm modification, days to silking and days to anthesis, anthesis-silking interval, number of kernels per row, number of rows per ear and grain yield. General analyses for the incomplete lattice block design and randomised complete block design for hybrid and inbred data respectively were performed using JMP 10 statistical software. Means were separated using the Tukey's test. Genetic analyses of data for grain yield and agronomic traits were conducted using a fixed effects model using REML in JMP 10. The importance of both GCA (51 percent) and SCA (49 percent) was observed for grain yield. A preponderance of GCA existed for ear height, days to anthesis, anthesis-silking interval, ears per plant and number of kernels per row, indicating that predominantly, additive gene effects controlled hybrid performance under optimum field conditions. The highest heritability was observed for days to silking (48.27 percent) suggesting that yield could be improved through selection for this trait. Under field conditions, variation in time to maturity was observed. This implies that these inbred lines can be recommended for utilisation in different agro-ecologies. Early maturing lines such as L18 can be used to introduce earliness in local cultivars, while early maturing single crosses such as L18 x L2, L5 x L9, L3 x L4 and L2 x L21 could be recommended for maize growers in drought prone areas such as the former Ciskei. Single crosses L18xL11, L16xL18, L8xL21 and L9xL6 had good tolerance to seedling drought stress. On the other hand, single crosses L18xL11 and L11xL13 had high grain yield and good endosperm modification. All these single crosses could be recommended for commercial production after evaluation across locations in the Eastern Cape Province. Alternatively they can be crossed with other superior inbreds to generate three or four way hybrids, which could then be evaluated for potential use by farmers in the Eastern Cape.
- Full Text:
- Date Issued: 2012
Clustering algorithms and their effect on edge preservation in image compression
- Authors: Ndebele, Nothando Elizabeth
- Date: 2009
- Subjects: Image compression , Vector analysis , Cluster analysis , Cluster anaylsis -- Data processing , Algorithms
- Language: English
- Type: Thesis , Masters , MSc
- Identifier: vital:5576 , http://hdl.handle.net/10962/d1008210 , Image compression , Vector analysis , Cluster analysis , Cluster anaylsis -- Data processing , Algorithms
- Description: Image compression aims to reduce the amount of data that is stored or transmitted for images. One technique that may be used to this end is vector quantization. Vectors may be used to represent images. Vector quantization reduces the number of vectors required for an image by representing a cluster of similar vectors by one typical vector that is part of a set of vectors referred to as the code book. For compression, for each image vector, only the closest codebook vector is stored or transmitted. For reconstruction, the image vectors are again replaced by the the closest codebook vectors. Hence vector quantization is a lossy compression technique and the quality of the reconstructed image depends strongly on the quality of the codebook. The design of the codebook is therefore an important part of the process. In this thesis we examine three clustering algorithms which can be used for codebook design in image compression: c-means (CM), fuzzy c-means (FCM) and learning vector quantization (LVQ). We give a description of these algorithms and their application to codebook design. Edges are an important part of the visual information contained in an image. It is essential therefore to use codebooks which allow an accurate representation of the edges. One of the shortcomings of using vector quantization is poor edge representation. We therefore carry out experiments using these algorithms to compare their edge preserving qualities. We also investigate the combination of these algorithms with classified vector quantization (CVQ) and the replication method (RM). Both these methods have been suggested as methods for improving edge representation. We use a cross validation approach to estimate the mean squared error to measure the performance of each of the algorithms and the edge preserving methods. The results reflect that the edges are less accurately represented than the non - edge areas when using CM, FCM and LVQ. The advantage of using CVQ is that the time taken for code book design is reduced particularly for CM and FCM. RM is found to be effective where the codebook is trained using a set that has larger proportions of edges than the test set.
- Full Text:
- Date Issued: 2009
- Authors: Ndebele, Nothando Elizabeth
- Date: 2009
- Subjects: Image compression , Vector analysis , Cluster analysis , Cluster anaylsis -- Data processing , Algorithms
- Language: English
- Type: Thesis , Masters , MSc
- Identifier: vital:5576 , http://hdl.handle.net/10962/d1008210 , Image compression , Vector analysis , Cluster analysis , Cluster anaylsis -- Data processing , Algorithms
- Description: Image compression aims to reduce the amount of data that is stored or transmitted for images. One technique that may be used to this end is vector quantization. Vectors may be used to represent images. Vector quantization reduces the number of vectors required for an image by representing a cluster of similar vectors by one typical vector that is part of a set of vectors referred to as the code book. For compression, for each image vector, only the closest codebook vector is stored or transmitted. For reconstruction, the image vectors are again replaced by the the closest codebook vectors. Hence vector quantization is a lossy compression technique and the quality of the reconstructed image depends strongly on the quality of the codebook. The design of the codebook is therefore an important part of the process. In this thesis we examine three clustering algorithms which can be used for codebook design in image compression: c-means (CM), fuzzy c-means (FCM) and learning vector quantization (LVQ). We give a description of these algorithms and their application to codebook design. Edges are an important part of the visual information contained in an image. It is essential therefore to use codebooks which allow an accurate representation of the edges. One of the shortcomings of using vector quantization is poor edge representation. We therefore carry out experiments using these algorithms to compare their edge preserving qualities. We also investigate the combination of these algorithms with classified vector quantization (CVQ) and the replication method (RM). Both these methods have been suggested as methods for improving edge representation. We use a cross validation approach to estimate the mean squared error to measure the performance of each of the algorithms and the edge preserving methods. The results reflect that the edges are less accurately represented than the non - edge areas when using CM, FCM and LVQ. The advantage of using CVQ is that the time taken for code book design is reduced particularly for CM and FCM. RM is found to be effective where the codebook is trained using a set that has larger proportions of edges than the test set.
- Full Text:
- Date Issued: 2009
- «
- ‹
- 1
- ›
- »