الفهرس 
Basic ConceptsOnly 14 pages are availabe for public view
 |  | from | 125 |  |  |
| | | from | 125 | | |
Abstract
Plants play an important role in our life. Without plants, there will be no existence of the earth’s environment. Unfortunately in recent days, many types of plants are at risk of extinction. To protect plants a plant database is an important step towards the conservation of the earth’s biosphere. But the huge number of plant species worldwide with massive features needs fast computing solution. For handling such volumes of information, the development of a quick and efficient classification method has become an area of active research. Sparse representation for images is used for recognition in the last years in a wide range.
On the other hand, applications of digital image processing for leaf recognition processing typically require massive computation power as the information required to be processed is vast. Thus, parallel algorithms are the solution that grants the capability to deal with huge datasets of plant leaves.
Some motivations of this thesis are:
• Providing a comparison between recent leaf recognition methods.
• Designing and implementing a new parallel effective leaf recognition application.
• Providing a comparison between the leaf recognition systems working sequentially against in parallel.
In this thesis, we proposed parallel leaf recognition systems based on sparse representation and morphological features. It implicates new parallel algorithms for accomplishing different processes for leaf recognition. It utilizes the computation ability of GPUs to recognize leaves rapidly and accurately.
This thesis consists of six chapters:
Chapter one provides an introduction to the thesis. It contains definitions for plants, some known properties for leaf features. It also presents the concepts of parallel computing, the parallel models, GPUs, and Multi-cores.
Chapter two presents various leaf datasets and recognition performance measurements. In addition, it presents a survey for plant various leaf recognition systems. The survey is classified into four groups: researches using shape features extraction for plant recognition; color and texture feature extraction; other various features; and some parallel methods for recognition.
In chapter three, we propose two modern plant leaf recognition system technologies. The first is the feature dictionary technique which is based on a morphological features extraction vector. The feature vector (FV) has 19 values including 17 shape features and two features for color and texture. Using four basic geometric features, we derive the rest of the shape features as leaf length, leaf width, leaf area, and leaf perimeter features, and other features’ values are deduced consequently. The second technique is based on sparse representation for leaf extracted patches. The two techniques are tested on two different leaf datasets, the Flavia and the Folio.
In chapter four, we present an accelerated plant recognition system, by improving the sparse representation technique to parallel implementation. The parallel proposed system tested on the Flavia dataset images with 1200×1600 pix dimension and patches size 60x60 pix. Also tested on the Leafsnap dataset with 600 x800 pix and patches size 60x60 pix.
In chapter five, we utilize the accuracy and fast execution achieved by the parallel recognition system and direct to powerful improvement to implement it on GPU. Experiments show that the execution time using GPU is accelerated by 4 comparisons by the previous execution and accuracy 99%.
Chapter six concludes all techniques presented in the thesis and suggests some ideas on the future work.