Selection of relevant features for classification from a high dimensional data set by keeping their class discriminatory information intact is a classical problem in Machine Learning. The classification power of the features can be measured from the point of view of redundant information and correlations among them. Choosing minimal set of features optimizes time, space complexity related cost and simplifies the classifier design, resulting in better classification accuracy. In this paper, tomato (Solanum Lycopersicum L) leaves and fruiting habits were chosen with a futuristic goal to build a prototype model of leaf & fruit classification. By applying digital image processing techniques, tomato leaf and fruit images were pre-processed and morphological shape based features were computed. Next, supervised filter and wrapper based feature selection techniques were adopted to choose the optimal feature set leading to small within-class variance and large among-class distance which may be of utter importance in building the model for recognition system of the tomato leaf and fruiting habit genre.