Vegetative indices and edge texture based shadow elimination method for rice plant images

Eliminating Shadows from the acquired images is an important preprocessing task for image analysis. In the work a novel shadow removal technique has been developed to eliminate shadows from the images acquired from the rice fields. Since the images are from the agricultural field, to remove shadow, vegetative indices, Normalized Difference Greenness Index (NDGI) and Redness Index (RI) of the image are calculated, based on which the infected (high RI) and uninfected (high NDGI) part of the image are identified. However, with ageing change of colour occurs in the plant when infected by diseases, due to which regions are wrongly identified using NDGI or RI index alone. So these two index values are combined to obtain the plant region (PR). However, the resultant image may contain shadow and background pixels due to cylindrical shape of the stem. To eliminate such pixels from the plant region, textures of the strips are utilized since shadow and background regions do not contain any strips. Edges of the image are determined using canny edge detection algorithm, where the edge detected image contains no shadow and background information. Median filter is applied to eliminate noisy edges and edge texture feature (ETF) is computed to obtain continuous edges in the image. Finally, shadow free image is obtained by applying AND operation between PR and ETF images.