| Summary: | Plant species identification and classification based on leaf shape is an important study in image processing. This is because each leaf carries substantial information that can be used to identify and classify the type of a plant. However, the task is complex due to existence of noise in the features of a leaf during image acquisition that can be influenced by other leaves that have similar features but with different categories or classes. To overcome this problem, an improved strategy has been conducted that includes pre-processing stage and mean feature extraction. This study presents the most popular statistical operators such as Mean Filtering Technique (MFT), Median Filtering Technique (MDFT), Adaptive (Wiener) Filtering Technique (WFT), Rank Order Filtering Technique (ROFT) and Adaptive Two-Pass Rank Order Filtering Technique (ATRFT) for noise removal during the pre-processing stage. Five different filtering techniques were applied to various categories or classes of plant leaf and their performance was evaluated using Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR). Ten morphological features were extracted from the pre-processed images. The mean values were calculated from the extracted features using Modified Weighted Mean approach (MWM). The retrievals and identification accuracy was evaluated using Precision and Recall measurements. The Wu's Standard database and the DCI database were used to test the proposed framework and the results were compared. The results showed the best filtering technique gives the higher identification performance for both of the databases. It is found that WFT is the best filtering technique and gives the best identification accuracy of 95.1% for Wu's Standard database. As for the DCI database, ATRFT presents the best filtering technique and gives the best identification performance of 83.1%. Based on the experimental results, the Wu's Standard database provided best results due to factors of less noise, occlusion and less distortion of some part of the leaves images than the DCI database during image acquisition. The major contribution of this study is in enhancing pre-processing stage which leads to the better plant species identification which is comparable with previous research. In conclusion, a better strategies in improving image pre-processing and mean feature extraction technique offers good result for the performance of shape-based image retrieval system as proven in this study.
|
|---|