Abstract-The this experiment is to identify the proper descriptor

Abstract-The Increase in number of diabetic
patients in day to day life due to lack of proper diet tends us to sought out a
solution for it. Food Image
Classification is a technique where a classifier assigns to the image one class out of a pre-defined
set of food classes, a set of characteristics representing the visual content
of the image is extracted and quantified. We use Bag-of-Features model
for automatic selection of food images, BoF methods are based on the order less collections of quantized
local image descriptors, they discard spatial information and are conceptually
and computationally simpler than many alternative methods. Image classification
undergoes key point extraction, builds a visual dictionary by using k-means
clustering and then finally classifies the food image into various classifiers.

Keywords-Bag-of-features (BoF), k-means clustering, Image classification, Key point

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now


Tremendous increase in number of diabetic patients worldwide, due to their
proven inability to assess their diet properly, had raised the need to develop
systems that will support type 1 diabetic (T1D) patients during CHO counting.
The increasing recent advances made in computer vision, permitted the
introduction of image analysis-based applications for diet management. In a
typical scenario, the user acquires an image of the upcoming meal using the
camera of his phone. The image is processed either locally or on the server
side in order to extract a series of features describing its visual properties.
The extracted features are fed to a classifier to recognize the various food
types of the acquired image, which will then be used for the CHO estimation.

scope of this experiment is to identify the proper descriptor size or
combination of sizes that should be used to describe the best performing key
point extraction technique. To this end, different sizes were evaluated and
then combined into a multi-scale scheme using a dense sampler. The used
descriptor sizes were 16, 24, 32 and 56 all their combinations with spacing
among them equal to 1/2 of each size in order to guarantee a sufficient number
of patches. The existing image analysis context, an
image is represented by the histogram of visual words, which are defined as
representative image patches of commonly occurring visual patterns. The BoW model is a simplifying representation used in
natural language processing and information retrieval. One aim of BoW is to categorize the
documents, which ignores the order
of the words belonging to a previously defined word dictionary and considers
only how frequently they appear. The concept of the BoF model adequately fits the food recognition
problem, since a certain food type is usually perceived as an Ensemble of
different visual elements mixed with specific proportions, but without any
typical spatial arrangement.

Existing Technique: BoW

Key points extraction is not such
easy task and also the data size and complexity in content are the drawbacks of
the existing system.


We propose a benchmark of several
objective functions for large-scale image classification1. Image category
selection is important to access visual information on the level of objects and
scene types2. The local descriptors are hierarchically quantized in a
vocabulary tree which allows a larger discriminatory vocabulary to used
efficiently3. Recently Bag-of-Features model had became more popular for
content based image classification with better performance and simplicity. We treat
images as collections of independent patches and then sampling the set of
patches. Evaluating the visual dictionary for each individual path
independently and using the resulting samples in Image classification4. We
use color histogram and bag of SIFT features to discriminate classifier5. Recently
a bag-of-features model was introduced into the area of computer vision as a
global image descriptor for difficult classifications6. BoF was derived from
the bag-of-words (BoW) model. BoW model is a popular way of representing
documents in natural language processing7.. The concept of BoF model
adequately fits the food selection problem 
because a certain food item is usually perceived as an ensemble of
different visual elements mixed with specific proportions without any spatial
arrangement7. A visual dataset with nearly 5000 homemade food images was
created, reflecting the nutritional habits8. A food selection strategy was
explained for the classification of image9.



image classification stage is involved in both training and testing phases. In
order to identify the appropriate classifier for the specific problem.
Identifying the appropriate descriptor size and type for a recognition problem
is a challenging task that involves a number of experiments. Its Inability to
capture any colour information constitutes a problem for the description of
many Objects, including foods. We analyse the problem of clustering food image
data list based RGB values

Algorithm: k-means clustering, the k-means
clustering is a iterative algorithm in which objects are moved among set of
clusters until the desired set is achieved. It is most popular and commonly
used method. This algorithm is well known for its high computational
complexity. The algorithm is built on the concept of user specified input parameter
(k). A set of n objects are divided into k clusters by the algorithm. A high
degree of similarity among elements in clusters is obtained.

Input: A data set
containing n objects, number of desired clusters k.

Output: A set consisting
of k clusters

Procedure: Initially
select k objects from the dataset as initial cluster centres. Assign each
object to the cluster with the nearest seed points, i.e., the centre points of
the cluster by using the Euclidean function. calculate the mean of
the all objects in the cluster. Repeat the process until the same points are assigned to each cluster in iterative rounds.

Advantages of k-means is that the Decision trees and neural networks were helpful to generate binary
classifiers of images and its time complexity.