Suitability of curvature as a feature for image-based pattern recognition: a case study on leaf image classification based on machine learning

Many features of leaves, including color, texture, and shape, are used in automated leaf recognition models. The curvature of a leaf is one of the least studied characteristics. This is primarily due to the fact that curvature is not an invariant feature and can fluctuate significantly in a single leaf in its many peripheral positions. The second reason is that if one considers curvature in a pixel-by-pixel manner, it seems entirely inappropriate as a single representative feature. In this study, we focused mostly on curvature and talked about how curvature might be used as a distinguishing feature for identifying leaves. We have provided a step-by-step method for dealing with curvature, starting with the fundamentals like the average curvature of a leaf and working our way up to a fine tuned representation that may be used as a significant feature. To determine whether the new feature is appropriate, we added it to the existing feature sets and fed the data to various standard classification models. The findings show a noticeable improvement in accuracy, demonstrating the usefulness of curvature as a feature.