This paper proposed a method that combines Polar Fourier Transform, color moments, and vein features to retrieve leaf images based on a leaf image. The method is very useful to help people in recognizing foliage plants. Foliage plants are plants that have various colors and unique patterns in the leaf. Therefore, the colors and its patterns are information that should be counted on in the processing of plant identification. To compare the performance of retrieving system to other result, the experiments used Flavia dataset, which is very popular in recognizing plants. The result shows that the method gave better performance than PNN, SVM, and Fourier Transform. The method was also tested using foliage plants with various colors. The accuracy was 90.80% for 50 kinds of plants.
Several researches in leaf identification have been explored, but there are still many challenges for researchers to try other approaches for better performance of the identification system. A certain method may give good performance in specific samples of leaves, but does not guarantee to perform good result for other ones. For example, ignoring colors in recognizing a foliage leaf is too risky. Sometime it is found that two or more plants have leaves with similar or same shape, but different colors. In that case, color features cannot be neglected. Therefore, a proposed method was accomplished to retrieve plants based on a leaf query using combination features: Polar Fourier Transform (PFT), colour moments, and vein of leaf. PFT is used to handle the shape of the leaf, colour moments are used to capture color information, and features extracted from the leaf's vein are used to improve performance of the retrieval system. This approach can be used to identify a leaf and also give top five of plants that have similar properties to the leaf query.
In order to assess the performance of the system, a leaf plant dataset came from Wu et al. [1] had been used. The result shows that the combination of PFT, colour moments, and vein of leaf features gives better performance than using SVM, PNN, and Fourier Moment [2]. The same method was also tested using 50 kinds of foliage plants that contain various colors. The average accuracy is 90.80%.
The remainder is organized as follows: Section 2 discusses related works, Section 3 describes all features used in the research, Section 4 explains how the mechanism of experiments is accomplished, Section 5 presents the experimental results, and Section 6 concludes the results.
Several researches in plant identification are described here. Warren [3] create a system that can measure the length and width of Chrysanthemum leaf and assesses several descriptive characters including the shape of the leaf apex, the shape of the base, the degree to which the margin is serrated and the depth and shape of the leaf’s lower sinus. Wu et al. [4] identified 6 species of plants. No color information was processed. They used aspect ratio, leaf dent, leaf vein, and invariant moment to identify plant. Wang et al. [5] used centroid-contour distance as shape features. Du et al. [6] captured the leaf shape polygonial approximation and algorithm called MDP (modified dynamic programming) for shape matching. Wu et al. [1] proposed an algorithm for plant recognition using Probabilistic Neural Network (PNN). They used 12 geometric features of leaf as input of identification system. By using 32 kinds of plants, the system has average accuracy 90,312%. The PNN algorithm is very fast to identify a leaf. However, manual processing should be done to locate the two terminal points of the leaf. Other researches, such as Singh et al. [2] proposed Support Vector Machine (SVM) to improve the performance of the indentification system, based on data came from Wu et. al [1], and Zulkifli [7] worked on 10 kinds of leaves and uses invariants moments as features to recognize them. The previous researchers did not incorporated color features in recognizing plants, but Man et al. did [8]. Man et. al. used features came from color and texture features and utilized SVM for classification. The accuracy of their system is 92% for 24 categories.
Other features that are incorporated in recognizing plants are extracted from vein of leaf (venation). Nam et al. [9] used shape and venation for leaf image retrieval. They used MPP (Minimum Perimeter Polygons) algorithm to solve the shape of the leaf and the type of venations as venation features. Li et al. [10] used ICA (Independent Component Analysis) to extract leaf vein. Other approach has done by Wu et al. [1], by performing morphological opening on grayscale image with flat, disk-shaped structuring element of radius 1, 2, 3, and 4. Then, by subtracting remaining image by the margin, the appearance of vein like image is obtained. That operation is quite simple and fast.
Meanwhile, Polar Fourier Transform (PFT) has been introduced in [11] to recognize 52 kinds of foliage plants. Compared to other methods (moment invariants and Zernike moments), PFT is prospective for recognizing shape of plants. However, using PFT only for foliage plant retrieval is not enough. Foliage plants mean that plants have leaves with unique shape, fancy pattern, or attractive colors. Therefore, handling color for features is a must, because some leaves of different species have same patterns but different colors.
Fourier transform (FT) is very populer in image processing, especially for analyzing purpose. The advantage of analyzing image in spectral domain over analyzing shape in spatial domain is that it is easy to overcome the noise problem which is common to digital images [12]. However, direct applying 2-D FT on a shape image in Cartesian space to derive feature descriptors is not practical du
This content is AI-processed based on open access ArXiv data.
