Abstract:Prototypical-part methods, e.g., ProtoPNet, enhance interpretability in image recognition by linking predictions to training prototypes, thereby offering intuitive insights into their decision-making. Existing methods, which rely on a point-based learning of prototypes, typically face two critical issues: 1) the learned prototypes have limited representation power and are not suitable to detect Out-of-Distribution (OoD) inputs, reducing their decision trustworthiness; and 2) the necessary projection of the learned prototypes back into the space of training images causes a drastic degradation in the predictive performance. Furthermore, current prototype learning adopts an aggressive approach that considers only the most active object parts during training, while overlooking sub-salient object regions which still hold crucial classification information. In this paper, we present a new generative paradigm to learn prototype distributions, termed as Mixture of Gaussian-distributed Prototypes (MGProto). The distribution of prototypes from MGProto enables both interpretable image classification and trustworthy recognition of OoD inputs. The optimisation of MGProto naturally projects the learned prototype distributions back into the training image space, thereby addressing the performance degradation caused by prototype projection. Additionally, we develop a novel and effective prototype mining strategy that considers not only the most active but also sub-salient object parts. To promote model compactness, we further propose to prune MGProto by removing prototypes with low importance priors. Experiments on CUB-200-2011, Stanford Cars, Stanford Dogs, and Oxford-IIIT Pets datasets show that MGProto achieves state-of-the-art image recognition and OoD detection performances, while providing encouraging interpretability results.

An Mlp-Orthogonal Gaussian Mixture Model Hybrid Model for Chinese Bank Check Printed Numeral Recognition

Offline Handwritten Numeral Recognition Using the Orthogonal Gaussian Mixture Model

Recognition of a Limited Chinese Character Set Based on PCA Learning Subspace Algorithm

Banknote image recognition algorithm based on multi-angle gaussian mixture model

Offline handwritten character recognition based on discriminative training of orthogonal Gaussian mixture model

A Hybrid Method For Robust Car Plate Character Recognition

Handwritten Character Recognition Using Semi-tied Full Covariance Gaussian Mixture Model

Paper Currency Recognition Using Gaussian Mixture Models Based On Structural Risk Minimization

The Design of HMM-based Banknote Recognition System

A Rejection Model Based on Multi-Layer Perceptrons for Mandarin Digit Recognition

Chinese Handwritten Legal Amount Recognition with HMM-Based Approach

A New Hybrid-Parameter Recurrent Neural Network for Online Handwritten Chinese Character Recognition

A new hybrid discriminative/generative model using the full-covariance multivariate generalized Gaussian mixture models

A Hybrid Post-Processing System For Offline Handwritten Chinese Character Recognition Based On A Statistical Language Model

Robust Math Formula Recognition in Degraded Chinese Document Images

Training Multilayer Perceptron with Multiple Classifier Systems

Multi-model Segmentation and Recognition Algorithm of Courtesy Amount

Mixture of Gaussian-distributed Prototypes with Generative Modelling for Interpretable and Trustworthy Image Recognition

A Hybrid Character Recognition Approach Using Fuzzy Logic and Stroke Bayesian Program Learning with Naïve Bayes in Industrial Environments

Dual Model Medical Invoices Recognition.

An MQDF-CNN Hybrid Model for Offline Handwritten Chinese Character Recognition.