| Summary: | Plant species identification based on leaf character is an important study in computer vision application. There is limited study on modeling for Ficus deltoidea varies recognition. Currently, one existing model is found and it is only valid within a fewer variety of F. deltoidea. the task becomes more complex in dealing with a wider variety of F. deltoidea since strong leaf features are highly diversed. The objective of this research is to develop an improved statistical recognition model for an efficient Ficus deltoidea Jack (Moraceae) varieties identification. The developed model consisted of three stages: (i) image pre-processing, (ii) measurement, representation and classification of leaf features, and (iii) parallelization. In image pre-processing stage, it utilized histogram shape-based thresholding and automatic petiole removal techniques to process the digital leaf images. The second stage is to provide the most relevant digital leaf features, to give a better feature representation and to select an appropriate classifier. This stage is accomplished by implementing several supplementary leaf features, a hybridization of filter and wrapper approach and the 1-Nearest neighbor algorithm. Lastly, parallelization stage is employed to increase the efficiency of the image processing (image pre-processing and feature measurement) phase. The developed parallel processing architecture consisted of four processor units under coarse-grain paradigm. A total of 420 F. deltoidea leaf images of six varieties are collected, trained, and tested. The experimental results are: (i) the digital leaf images have minimal segmentation errors (<1%) on several ground truth images for the proposed image pre-processing model, (ii) the recognition accuracy results indicate that the developed statistical model is increased up to 1 to 9% and (iii) the parallel model achieved the speedup factor by at least 2.68 times higher using four processor units as compared to sequential algorithm. In conclusion, a better statistical recognition model for F. deltoidea varieties identification by improving the above-mentioned stages offers good segmentation results, higher recognition accuracy and increase the performance of the processing is successfully developed.
|
|---|