In this work, we investigate several methods and strategies to learn deep embeddings for face recognition, using joint sample- and set-based optimization. We explain our framework that expands traditional learning with set-based supervision together with the strategies used to maintain set characteristics. We, then, briefly review the related set-based loss functions, and subsequently we propose a novel Max-Margin Loss which maximizes maximum possible inter-class margin with assistance of Support Vector Machines (SVMs). It implicitly pushes all the samples towards correct side of the margin with a vector perpendicular to the hyperplane and a strength inversely proportional to the distance to it. We show that the introduced loss outperform the previous sample-based and set-based ones in terms verification of faces on two commonly used benchmarks.
We propose novel color-blob-based COSFIRE filters.
They are effective for recognizing also objects with diffuse region boundaries.
Such a filter models (a part of) an object by a specific arrangement of color blobs.
The blobs contain information about the sizes and colors of the interior of regions.
We achieve high recognition rates: GTSRB (98.94%) and Butterfly (89.02%) data sets.
Most object recognition methods rely on contour-defined features obtained by edge detection or region segmentation. They are not robust to diffuse region boundaries. Furthermore, such methods do not exploit region color information. We propose color-blob-based COSFIRE (Combination of Shifted Filter Responses) filters to be selective for combinations of diffuse circular regions (blobs) in specific mutual spatial arrangements. Such a filter combines the responses of a certain selection of Difference-of-Gaussians filters, essentially blob detectors, of different scales, in certain channels of a color space, and at certain relative positions to each other. Its parameters are determined/learned in an automatic configuration process that analyzes the properties of a given prototype object of interest. We use these filters to compute features that are effective for the recognition of the prototype objects. We form feature vectors that we use with an SVM classifier. We evaluate the proposed method on a traffic sign (GTSRB) and a butterfly data sets. For the GTSRB data set we achieve a recognition rate of 98.94%, which is slightly higher than human performance and for the butterfly data set we achieve 89.02%. The proposed color-blob-based COSFIRE filters are very effective and outperform the contour-based COSFIRE filters. A COSFIRE filter is trainable, it can be configured with a single prototype pattern and it does not require domain knowledge.
The most frequent non-skin cancer type is breast cancer which is also named one of the most deadliest diseases where early and accurate diagnosis is critical for recovery. Recent medical image processing researches have demonstrated promising results that may contribute to the analysis of biopsy images by enhancing the understanding or by revealing possible unhealthy tissues during diagnosis. However, these studies focused on well-annotated and -cropped patches, whereas a fully automated computer-aided diagnosis (CAD) system requires whole slide histopathology image (WSI) processing which is, in fact, enormous in size and, therefore, difficult to process with a reasonable computational power and time. Moreover, those whole slide biopsies consist of healthy, benign and cancerous tissues at various stages and thus, simultaneous detection and classiffication of diagnostically relevant regions are challenging.
We propose a complete CAD system for efficient localization and classification of regions of interest (ROI) in WSI by employing state-of-the-art deep learning techniques. The system is developed to resemble organized work ow of expert pathologists by means of progressive zooming into details, and it consists of two separate sequential steps: (1) detection of ROIs in WSI, (2) classification of the detected ROIs into five diagnostic classes. The novel saliency detection approach intends to mimic efficient search patterns of experts at multiple resolutions by training four separate deep networks with the samples extracted from the tracking records of pathologists' viewing of WSIs. The detected relevant regions are fed to the classification step that includes a deeper network that produces probability maps for classes, followed by a post-processing step for final diagnosis.
In the experiments with 240 WSI, the proposed saliency detection approach outperforms a state-of-the-art method by means of both efficiency and eectiveness, and the final classification of our complete system obtains slightly lower accuracy than the mean of 45 pathologists' performance. According to the Mc- Nemar's statistical tests, we cannot reject that the accuracies of 32 out of 45 pathologists are not different from the proposed system. At the end, we also provide visualizations of our deep model with several advanced techniques for better understanding of the learned features and the overall information captured by the network