Color Image Processing of Weed Classification: A comparison of two Feature Extraction Technique

Abstract

It is a common practice in most image recognition applications to convert color images to gray scale prior to analysis. By doing so, the original colored image will lose its color features which may be critical especially when the recognition is based on the color information. It is not so critical if the image recognition task is based on other contexts such as shape or texture. In this paper, we address the issue of color image processing with specific application on weed classification task. The task of classifying weed according to its classes as either narrow or broad type can be best performed based on the texture context. This can be done either in gray scale or color (RGB) mode. Typically, the gray scale mode is adopted in which the original colored weed images of RGB values are converted to gray scale. In this work, we intend to show that by combining both color and texture information of the weed images, higher classification accuracy can be afforded. Color filtering algorithm was implemented in the preprocessing task involving the technique known as the extracted green color. Following the color filtering algorithm implementation, the images were subjected to two types of feature extraction techniques namely the FFT and gray level co-occurrence matrix (GLCM). Next, we perform classification and compare the performance of color based processing against grayscale processing using the GLCM and FFT based feature vectors.