| Popis: |
Applications of machine vision for automated inspection and sorting of fruits have been widely studied by scientists and engineers. In these applications, edge detection, segmentation, and shape recovery are difficult problem. Previous studies have usually adopted some preprocessing such as noise removal and motion deblurring before using a threshold method to detect shape boundary. In many cases, however, this manner is troubled and not unified and does not work well. This research proposes a novel approach for fruit shape detection in RGB spaces based on a fast level set method and the Chan-Vese model. We called it optimizing Chan-Vese model (OCV). This new algorithm is fast because it needs no re-initialization procedure and thus is suitable for fruit sorting. OCV has three advantages compared to traditional methods. First, it provides a unified framework for detection fruit shape boundary, requiring no preprocessing and even if the raw image is noisy or blurred. Second, it can detect boundaries for images of fruit with multi-colored edges, which traditional methods fail to deal with. Third, it is processed directly in colour space without any transformations that can lose much information. The proposed method has been applied to fruit shape detection with promising results. |
