Design and Development of an Image Based Search Engine for Plants

Abstract

he main objective of our research work was to develop an image based search newlineengine for plants. Its need was felt because of the huge diversity existing on our earth newlineand the large inter-species similarity which makes the manual process of recognition newlinedifficult and tedious. Among various parts of a plant, the leaf is easily available in all newlineseasons and can be easily scanned and studied as a 2D object. Almost most of the newlineleaves are green in color and the texture of the leaves is difficult to be captured. So, newlinethe shape of the leaf is considered for the identification of plants. Hence, for newlineautomating the process of recognition of plant using leaves, the key challenges newlineidentified were to keep the size of feature vector reasonable and still achieve accurate newlineresults. As the results generated by the search engine would be interpreted by the user newlinefor their similarity, the theories behind human perception for a leaf were understood newlineso as to reduce semantic gap. newlineThe Shape based Leaf Recognition System: SbLRS was designed and implemented newlineusing NetBeans IDE with Java as frontend and Oracle 10g as backend. SbLRS used newlinethree stage search procedure for the identification process. The important parameters newlineof a leaf shape were assigned weightage, depending upon their importance in defining newlinethe details. Quad Centroid Distance Variation (QCDV) was a novel method proposed newlineby us. This method overcame the drawback of large vector size of centroid distance newlineapproach and represented the curvature details of the leaf shape quadrant-wise. The newlineFirst level of SbLRS considered global vector , the second level used the leaf newlinemargin and the third level used the leaf angle for plant identification. The results newlinegenerated by SbLRS were shown with their degree of matching. The testing of SbLRS newlinewas been done on two database: UC Irvine Machine Learning Repository: Leaf and newlineUser-created Database. Various contour based image retrieval methods were newlineimplemented and the results generated were compared in terms of classification newlineaccuracy, recall and pr