Cervical cancer affects more than 0.5 million women annually causing more than 0.3 million deaths. Detection of cancer in its early stages is of prime importance for eradicating the disease from the patient’s body. However, regular population-wise screening of cancer is limited by its expensive and labour intensive detection process, where clinicians need to classify individual cells from a stained slide consisting of more than 100,000 cervical cells, for malignancy detection. Thus, Computer-Aided Diagnosis (CAD) systems are used as a viable alternative for easy and fast detection of cancer. In this paper, we develop such a method where we form an ensemble-based classification model using three Convolutional Neural Network (CNN) architectures, namely Inception v3, Xception and DenseNet-169 pre-trained on ImageNet dataset for Pap stained single cell and whole-slide image classification. The proposed ensemble scheme uses a fuzzy rank-based fusion of classifiers by considering two non-linear functions on the decision scores generated by said base learners. Unlike the simple fusion schemes that exist in the literature, the proposed ensemble technique makes the final predictions on the test samples by taking into consideration the confidence in the predictions of the base classifiers. The proposed model has been evaluated on two publicly available benchmark datasets, namely, the SIPaKMeD Pap Smear dataset and the Mendeley Liquid Based Cytology (LBC) dataset, using a 5-fold cross-validation scheme. On the SIPaKMeD Pap Smear dataset, the proposed framework achieves a classification accuracy of 98.55% and sensitivity of 98.52% in its 2-class setting, and 95.43% accuracy and 98.52% sensitivity in its 5-class setting. On the Mendeley LBC dataset, the accuracy achieved is 99.23% and sensitivity of 99.23%. The results obtained outperform many of the state-of-the-art models, thereby justifying the effectiveness of the same. The relevant codes of this proposed model are publicly available on GitHub.

Text non-text separation is one of the most essential pre-processing steps for any optical character recognition (OCR) system. As an OCR engine can only process texts, the non-texts present in an input document image are required to be suppressed at the initial level. Therefore, to build a complete OCR system, an efficient text non-text separation module needs to be developed. To this end, we have proposed a texture-based feature descriptor followed by a novel feature selection technique for region-based text non-text classification. First, we have incorporated rotation invariant property with local ternary pattern to form a new texture-based feature descriptor, rotation invariant local ternary pattern (RILTP). Next, a novel feature selection technique is proposed which is a modified version of binary particle swarm optimization (BPSO). For the evaluation of the proposed text non-text classification method, we have initially constructed a database consisting of 690 images of text and non-text regions extracted from 70 pages of RDCL 2015 and 75 pages of RDCL 2017 page segmentation competitions databases. In this database, each class contains 345 data samples. The proposed texturebased feature descriptor has obtained an accuracy of 97.09% on this database. Whereas, after applying BPSO, the feature dimension is reduced by approximately 55% and at the same time, the accuracy reaches 97.5%. Furthermore, in this work, another database is also created from Media team document pages to validate the robustness of this method. The second database comprises 100 text and 100 non-text images. The method has achieved 96.28% accuracy when it is trained with the first database and tested with the second database. The comparative study reveals the robustness and strength of the proposed method as it outnumbers many state-of-the-art texture-based features. Besides, the proposed feature selection method is also compared with various standard feature selection methods, and it has been observed that the proposed one outperforms all those methods considered here for comparison.