Cervical Cell Recognition based on the Leave-one-out Feature Selection Algorithm
Zhao Li, Na Dong, Jianfang Chang · 2019
In order to reduce the mortality rate of cervical cancer, it is of great significance to implement timely and effective screening of cervical cancer for women. Cervical cell recognition based on image processing and computer science has become a trend, however, most methods extract too many cell features, resulting in lower recognition efficiency. In order to obtain effective cell features and represent the highest accuracy with the least features, a cervical cell recognition model based on the leave-one-out feature selection algorithm is proposed. Firstly, the improved Canny algorithm is used to extract the contour of the cells, and then color, texture and morphological features are extracted. In order to avoid low efficiency caused by feature redundancy, this paper uses the leave-one-out method to select the extracted features. Finally, this paper uses Support Vector Machine (SVM) to construct the recognition model. The experimental results illustrate that the improved Canny algorithm proposed in this paper can effectively extract precise cell edges, so that more accurate cell features are obtained. Leave-one-out method can select more representative features, maintaining the recognition accuracy while effectively reducing its running time.