| 摘要 |
A method for automated selection of image regions (“tiles”) in an overall image includes computing a gray-level co-occurrence matrix and entropy of a green plane of the overall image, applying morphological closing to the matrix and thresholding the matrix and entropy image to provide binary masks. The matrix and entropy masks are combined with a vignette mask, the combination indicating areas of acceptable tissue from which tiles are selected randomly. For cancer grading; image data is transformed to Hue, Saturation and Value; for steroid/protein expression analysis it is transformed to cyan and a Sobel of cyan is computed. A feature measure is computed for each tile based on color and texture, and is carried out randomly but influenced by feature measure. Finally, from the further selection, tiles are chosen which combine high feature measure with low overlap. |
