Towards Personalised Training of Machine Learning Algorithms for Food Image Classification Using a Smartphone Camera

This work is related to the development of a personalised machine learning algorithm that is able to classify food images for food logging. The algorithm would be personalised as it would allow users to decided what food items the model will be able to classify. This novel concept introduces the idea of promoting dietary monitoring through classifying food images for food logging by personalising a machine learning algorithm. The food image classification algorithm will be trained based on specific types of foods decided by the user (most popular foods, food types e.g. vegetarian). This would mean that the classification algorithm would not have to be trained using a wide variety of foods which may lead to low accuracy rate but only a small number of foods chosen by the user. To test the concept, a range of experiments were completed using 30 different food types. Each food category contained 100 images. To train a classification algorithm, features were extracted from each food type, features such as SURF, LAB colour features, SFTA, and Local Binary Patterns were used. A number of classification algorithms were used in these experiments; Nave Bayes, SMO, Neural Networks, and Random Forest. The highest accuracy achieved in this work was 69.43 % accuracy using Bag-of-Features (BoF) Colour, BoF-SURF, SFTA, and LBP using a Neural Network.

[1]  J. A. Hartigan,et al.  A k-means clustering algorithm , 1979 .

[2]  Keiji Yanai,et al.  A food image recognition system with Multiple Kernel Learning , 2009, 2009 16th IEEE International Conference on Image Processing (ICIP).

[3]  Luc Van Gool,et al.  SURF: Speeded Up Robust Features , 2006, ECCV.

[4]  Matti Pietikäinen,et al.  Performance evaluation of texture measures with classification based on Kullback discrimination of distributions , 1994, Proceedings of 12th International Conference on Pattern Recognition.

[5]  Keiji Yanai,et al.  FoodCam-256: A Large-scale Real-time Mobile Food RecognitionSystem employing High-Dimensional Features and Compression of Classifier Weights , 2014, ACM Multimedia.

[6]  Maurice K. Wong,et al.  Algorithm AS136: A k-means clustering algorithm. , 1979 .

[7]  J. Platt.  Sequential Minimal Optimization : A Fast Algorithm for Training Support Vector Machines , 1998 .

[8]  Giovanni Maria Farinella,et al.  A Benchmark Dataset to Study the Representation of Food Images , 2014, ECCV Workshops.

[9]  Edward J. Delp,et al.  Food texture descriptors based on fractal and local gradient information , 2011, 2011 19th European Signal Processing Conference.

[10]  Huiru Zheng,et al.  Semi-automated system for predicting calories in photographs of meals , 2015, 2015 IEEE International Conference on Engineering, Technology and Innovation/ International Technology Management Conference (ICE/ITMC).

[11]  Peter Scarborough,et al.  The economic burden of ill health due to diet, physical inactivity, smoking, alcohol and obesity in the UK: an update to 2006-07 NHS costs. , 2011, Journal of public health.

[12]  Matthieu Guillaumin,et al.  Food-101 - Mining Discriminative Components with Random Forests , 2014, ECCV.

[13]  Leo Breiman,et al.  Random Forests , 2001, Machine Learning.

[14]  Edward J. Delp,et al.  Combining global and local features for food identification in dietary assessment , 2011, 2011 18th IEEE International Conference on Image Processing.

[15]  Keiji Yanai,et al.  FoodCam: A Real-Time Mobile Food Recognition System Employing Fisher Vector , 2014, MMM.

[16]  Agma J. M. Traina,et al.  An Efficient Algorithm for Fractal Analysis of Textures , 2012, 2012 25th SIBGRAPI Conference on Graphics, Patterns and Images.