We propose a dynamic thresholding method with feature-feedback in which feature information of the thresh-olded result is used to tune the thresholding in a proper state where a optimal thresholded result can be obtained. An original image is convolved with a two-dimensional Gaussian function first, and then the convolved image is used as a surface threshold at which the original image is thresholded. For the thresholded image, The selected feature value of extracted objects is calculated first and then, as the feedback information, is compared with the reference value. The space constant σ of the Gaussian is adjusted according to the error between the two values recursively until a proper result is given.
The proposed method is applied to extracting nuclei in the glomeruli of human kidneys, where two features are selected as the feedback information; one is the average size of the extracted nuclei, and another is their size-number distribution. Experimental results are presented to successfully demonstrate the feasibility of the proposed method.