No abstract

This work is based on color image segmentation by spatial-color pixel classification in Luv color space. Classes of pixels are difficult to be identified when the color distributions of the different objects highly overlap in the color space and when the color points give rise to non-convex clusters. It is proposed to apply spectral classification to regroup the pixels which represent the same regions, into classes. Spectral clustering achieves a spectral decomposition of a similarity matrix in order to construct an eigen-space in which the clusters are expected to be well separated. The similarity matrix used in this paper is derived from a spatial-color compactness function. This function takes into account both the distribution of colors in the color space and the spatial location of colors in the image plane. Spectral clustering that uses FCM performs better in Luv color space when compared with other Spectral clustering algorithms..

Color is an important feature in applications like the detection of plants and diseases in plants. Deep learning networks utilize optimizers towards improving the accuracy of classification. Color space is treated as an extra dimension with which the image could be better classified. Hence a particular blending of classifier, optimizer and color space is expected to provide enhanced accuracy of classification. There are very rare cases of studies having examined the effect of color space with deep learning networks. Hence, it is motivated to study the role of color spaces. Leaf datasets available in literature have been utilized. Of the few tried networks, Inception V3 is found to perform better with optimizer Adam. Color space XYZ performed better than RGB in the above combination. It has also been tried to obtain majority voting among various optimizer combinations. This solution is also better with XYZ color space. Among the various datasets utilized, consistent performance has been observed with Flavia data set yielding superior classification accuracy.