Ensemble of handcrafted and deep learned features for cervical cell classification


The aim of this work is to propose an ensemble of descriptors for Cervical Cell Classification. The system proposed here achieves strong discriminative power that generalizes well thanks to the combination of multiple descriptors based on different approaches, both learned and handcrafted. For each descriptor, a separate classifier is trained, then the set of classifiers is combined by sum rule. The system we propose here also presents a simple and effective method for boosting the performance of trained CNNs by combining the scores (using sum rule) of multiple CNNs into an ensemble. Different types of ensembles and different CNN topologies with different learning parameter sets are evaluated. Moreover, features extracted from tuned CNNs are used for training a set of Support Vector Machines (SVM). First, we validate our method on two cervical cell-related datasets; then, for more in-depth validation, we test the same system on other bioimage classification problems. Results show that the proposed system obtains state-of-the-art performance in all datasets, despite not being tuned on a specific dataset, i.e. the same descriptors with the same parameters are used in all the datasets. The MATLAB code of the descriptors will be available at https://github.com/LorisNanni.


Information Technology and Cybersecurity

Document Type





Bioimage classification, Cancer data analysis, Deep learning, Ensemble of classifiers

Publication Date


Journal Title

Intelligent Systems Reference Library