Acknowledgement
We would like to express our sincere gratitude to our project guide, Prof. Reena Saini, for her
continuous guidance, valuable suggestions, and constant support throughout the development
of our project “Food Image Recognition System.” Her encouragement and expert insights
greatly helped us understand the concepts of deep learning and complete this work successfully.
We are also thankful to the Department of Information Technology, BKBIET Pilani, for
providing the necessary facilities, resources, and a supportive academic environment. We
extend our appreciation to all faculty members for their consistent help and motivation.
We are deeply grateful to our parents, classmates, and friends for their encouragement and
support during the entire project period. Their cooperation played an important role in the
successful completion of this work.
Submitted by:
Kartik Jangir (22EBKIT011)
Manish Kumar (22EBKIT013)
Table of Contents
1. Acknowledgement
2. Introduction
3. Objective of the Project
4. Problem Statement
5. Proposed Solution
6. System Architecture
7. Technology Stack Used
8. Module Description
8.1. Dataset Module
8.2. Preprocessing Module
8.3. CNN Model Module
8.4. Training Module
8.5. Prediction Module
9. Dataset Schema
10. Workflow & Sequence Diagrams
11. Key Features of the System
12. Implementation Screenshots
13. Testing & Validation
14. Conclusion
15. Future Scope
16. References
Introduction
In today’s world, the use of artificial intelligence is growing in almost every field, and one
useful application is food image recognition. It is a technology that allows a computer to
identify food items just by looking at an image. This is helpful in many areas such as diet
tracking apps, restaurant automation, health monitoring, and smart kitchen systems. Instead of
manually entering what food a person is eating, the system can automatically recognize it from
a photo.
The main idea of this project is to build a Food Image Recognition System using Deep
Learning. We use a model that can learn features from images and classify them into different
food categories. For this purpose, Convolutional Neural Networks (CNNs) are used because
they are very effective for image-related tasks. They can understand patterns like color, shape,
texture, and details in food images.
The project is developed using Jupyter Notebook, which makes it easy to run code step-by-
step, view outputs instantly, and visualize the training results. Our system goes through
different stages like collecting the dataset, preprocessing the images, training the model, testing
its accuracy, and finally predicting the food item from new images.
Overall, this project shows how deep learning can be used to automatically identify food items,
making tasks easier and faster. It also provides a strong foundation for more advanced
applications such as calorie counting, nutrition analysis, food logging, and smart restaurant
systems.
Objective of the Project
The main aim of this project is to design a system that can automatically recognize different
types of food items from an image with the help of deep learning techniques. In today’s digital
world, food identification has become important for various applications such as diet
monitoring, mobile food apps, restaurant systems, and smart kitchen technologies. This project
focuses on making the recognition process simple, fast, and accurate.
The key objectives of the project are:
1. To develop an intelligent system for food classification
To build a model that can analyze a food image and correctly identify what type of food
it is, without any manual input.
2. To make use of deep learning for better understanding of images
To use Convolutional Neural Networks (CNNs), which are highly efficient for image-
based tasks, so the system can learn patterns and features from food images.
3. To prepare and process the food image dataset effectively
To resize, normalize, and augment the images so that the model learns well from
different variations of the same food.
4. To test the accuracy and reliability of the model
To evaluate how well the system performs on new food images and ensure the
predictions are consistent and reliable.
5. To create a foundation for real-world applications
To build a system that can later be expanded for advanced uses such as calorie detection,
restaurant billing, real-time camera identification, and mobile app integration.
Problem Statement
With the rapid growth of health-focused applications, food delivery platforms, and smart
automation systems, the need for accurate food recognition and nutritional analysis has become
increasingly important. However, manual identification of food from images is slow,
subjective, and unreliable. Many dishes look similar, and factors such as angle, lighting, and
presentation can make the recognition process even more challenging.
Furthermore, manually calculating nutrition content such as protein, fat, carbohydrates, and
vitamins for every food item is a difficult and time-consuming task. Most users are not familiar
with nutritional values, and estimating them without technical tools is almost impossible.
Traditional approaches require users to search online or consult nutritional charts, which is
inconvenient and impractical for everyday use.
Another major limitation of existing systems is the lack of automation. Users usually need to
enter food names manually, select categories, or estimate nutritional values themselves. These
methods are inefficient and do not provide accurate insights, especially when the person is
trying to maintain a diet, track calories, or monitor health.
Therefore, there is a strong need for a system that can:
• Automatically recognize the food item from an image,
• Display the uploaded picture,
• Show the probability of predictions,
• Accept the quantity of food,
• And provide a complete nutrition report with visual charts.
This project aims to solve this problem by developing a deep learning-based Food Image
Recognition and Nutrition Analysis System that uses Convolutional Neural Networks (CNNs)
for image classification and a Flask web interface for real-time predictions and pie chart
visualization. The system reduces human effort, improves accuracy, and provides users with
immediate nutritional information based on the recognized food item.
Proposed Solution
To solve the problem of manually identifying food items and calculating their nutrition, this
project provides an easy and automatic solution. The main idea is to create a system that can
look at a food image and recognize what food it is, without needing any human help. Instead
of the user typing the food name or guessing the nutrition values, the system does everything
automatically.
The proposed solution uses an image-based learning method that allows the computer to
understand and learn different types of food by looking at many examples. After learning, the
system becomes able to identify new food images. When the user uploads a picture of any food,
the system studies the image, compares it with what it has learned before, and predicts the most
likely food item. Along with the prediction, it also shows how confident the system is, using
simple percentage values.
To make the system more useful, it also includes a nutrition calculator. After the food is
recognized, the user can enter the quantity in grams. Based on this quantity, the system
calculates nutrition values such as protein, fat, carbohydrates, and other important nutrients.
These values are shown in a clear and simple form so that anyone can understand them easily.
A small chart is also displayed to help the user see the nutritional balance at a glance.
All steps—uploading the image, viewing the prediction, entering the quantity, and checking
the nutrition—are designed to be simple and comfortable for the user. The goal of the proposed
solution is to reduce human effort, avoid confusion, and make food recognition and nutrition
checking easier for everyone. This system can be very useful for diet tracking, health
monitoring, and general awareness about the food we eat.
System Architecture
The system architecture describes how the different parts of the project work together to
recognize food from an image and calculate its nutrition. The entire process is divided into a
few simple stages, and each stage has a specific role. These stages work in a step-by-step
manner to produce the final result for the user.
1. Image Input Stage
The first stage of the system is the image input. In this stage, the user selects and uploads
a picture of the food. The system receives this picture and prepares it for further steps.
This stage is important because the quality of the input image affects the accuracy of
the final prediction.
2. Image Processing Stage
Once the image is received, the system processes it by adjusting its size, shape, and
color format. This step ensures that all images look similar in form, so the learning
system can understand them easily. Processing the image also removes any unnecessary
background information and focuses on the food portion.
3. Learning and Recognition Stage
This is the core part of the system. Here, the image is sent to the trained learning model.
The model has already learned many food categories from a large number of example
images. When a new image is given, the model studies it and compares it to what it has
learned before. Based on this comparison, it predicts the most likely food item and also
gives a confidence percentage for the prediction.
4. Result Display Stage
After identifying the food, the system shows the result to the user. It displays the
uploaded image along with the predicted food name and the confidence score. This
helps the user understand what the system recognized and how sure it is about the
prediction.
5. Quantity Input Stage
In the next step, the user can enter the quantity of the food in grams. The system needs
this quantity to calculate the nutrition values correctly. Different amounts of food have
different nutrition levels, so this stage helps customize the result.
6. Nutrition Calculation Stage
Based on the recognized food and the quantity entered by the user, the system calculates
the nutrition values. This includes protein, fat, carbohydrates, vitamins, and other
nutrients. These values are taken from a stored nutrition database and multiplied
according to the amount of food. This stage makes the system useful for health tracking
and diet management.
7. Nutrition Visualization Stage
To make the information easy to understand, the system displays the nutrition values in
a simple chart. This visual chart helps the user quickly see the nutrient distribution in
the food portion.
8. Final Output Stage
In the final stage, the user receives all the information clearly: the recognized food item,
the prediction accuracy, the quantity-based nutrition values, and the nutrition chart. The
user can then take decisions based on this data or upload another image if needed.
Technology Stack Used
The technology stack explains all the tools, software, and resources used to build the Food
Image Recognition and Nutrition Analysis System. Each technology plays an important role in
making the system work smoothly and efficiently.
1. Programming Language
The entire system is developed using Python. Python is easy to understand, has a clean
syntax, and supports many useful libraries for image handling, data analysis, and
machine learning. Its flexibility makes it ideal for building projects that involve image
recognition.
2. Development Environment
The project is created and tested in Jupyter Notebook, which allows running code step-
by-step and viewing results immediately. It is simple to use and helps in understanding
how each part of the project works. Jupyter Notebook is also useful for visualizing
predictions and debugging errors easily.
3. Image Handling Libraries
To load, resize, and process images, the system uses image-related Python libraries.
These libraries make it easy to read food images, convert them into the required format,
and prepare them before sending them to the learning model.
4. Machine Learning Libraries
Python provides powerful libraries for building and training learning systems. These
libraries help the project learn from large food datasets and make predictions on new
images. They support many functions like model training, evaluation, and probability
prediction.
5. Data Analysis and Numerical Libraries
Libraries for handling numbers and data are used to store nutrition values, calculate
quantities, and display results. These libraries help perform mathematical operations
quickly and efficiently, especially when computing nutrition based on food quantity.
6. Graph and Chart Libraries
To show the nutritional breakdown in a simple and visual way, the system uses chart-
making libraries. These libraries help create pie charts, bar graphs, and other visual
elements that make the nutrition data easy to understand for users.
7. Supporting Tools
Additional tools such as file-handling modules and data-formatting libraries are used to
manage images, load nutrition datasets, and organize results. These tools make the
system more stable and easier to maintain.
Module Description
The system is divided into several important modules, and each module has a specific role in
completing the food recognition and nutrition analysis process. All these modules work
together in a smooth flow, starting from taking the food image to displaying the nutrition report.
1. Dataset Module
The Dataset Module manages all the images used for learning and testing the system. The food
images are stored in a well-organized structure, where each food category has its own folder.
This helps the system understand which image belongs to which class. The dataset contains
many samples so the model can learn different appearances of the same food, such as different
angles, lighting conditions, and presentation styles. The module also separates the data into
training and testing sets, ensuring that the system learns from one part of the data and is
evaluated on another.
2. Preprocessing Module
This module prepares all food images so that they can be used properly by the learning model.
Since the images come in different sizes and formats, they are resized to a fixed dimension and
normalized. This makes all images uniform and easy for the system to process. The module
may also apply simple transformations like rotation, zoom, or flipping to increase the variety
of training images. These steps help the system handle real-life variations and improve
accuracy.
3. CNN Model Module
The CNN Model Module is responsible for learning the visual features of each food item. It
contains multiple layers that analyze the image step-by-step. In the beginning, the model learns
simple patterns like edges and colors. Later, it identifies more complex features like texture,
shape, and food structure. By learning from thousands of examples, the model develops the
ability to recognize food items even if they appear slightly different. This module is the “brain”
of the system because it performs the main learning and classification tasks.
4. Training Module
In this module, the model is trained using the prepared dataset. The system looks at each image
and tries to guess the correct food category. If the guess is wrong, the model adjusts its internal
settings to improve the next prediction. This process is repeated for several rounds until the
model becomes accurate. The training module also records important values like accuracy and
loss to show how well the model is learning over time. The best-performing version of the
model is saved for later use.
5. Prediction Module
The Prediction Module is the final and most interactive part of the system. After the model has
been trained, this module uses it to identify new food images. When the user uploads a food
picture, the module processes it and predicts the most likely food item along with the
confidence score. It also displays the uploaded image for user confirmation. After predicting
the food type, the user can enter the quantity, and the module calculates the nutrition values
based on that amount. A simple chart is generated to visually show the nutritional composition.
Dataset Schema
The dataset used in this project is arranged in a clear folder structure so the system can easily
learn different categories of food images. The images are divided into two main parts: one for
training and one for testing.
1. Folder Structure
• The train section contains images used for teaching the model.
• The test section contains images used for checking the model’s performance.
• Each class has its own folder, and the folder name represents the category.
2. Image Format
• Images can be in common formats like JPG or PNG.
• Sizes may vary, but they are resized during preprocessing.
• Each class contains many images so the system can learn different patterns.
3. Nutrition Data Structure
The nutrition information is stored separately in a small table.
It includes nutrient values per 100 grams, such as:
• Energy
• Protein
• Fats
• Carbohydrates
• minerals or vitamins
• These values are later scaled according to the user’s entered quantity.
Workflow & Sequence Diagrams
The workflow of the system explains how the entire process works — from taking an image as
input to producing the final nutrition report. The sequence of steps is arranged in an easy-to-
understand order so that every part of the system works smoothly and correctly.
1. Workflow of the System
The system follows a step-by-step workflow. Each stage performs a specific task and passes
its output to the next stage.
Step 1: Image Input
The user starts by uploading an image of the food.
This image becomes the main input for the system.
Step 2: Image Processing
The uploaded image is resized and prepared so that it can be correctly analyzed.
This step helps remove unnecessary background details and keeps the image format
consistent.
Step 3: Food Recognition
After processing, the image is sent to the learning system.
Here, the system studies the image and predicts which food category it belongs to.
It also shows how confident it is in its prediction using a percentage score.
Step 4: Display of Results
Once the food type is identified, the result is displayed along with the confidence score and
the uploaded image.
This helps the user confirm that the system recognized the correct food.
Step 5: Quantity Input
The user enters the quantity of food in grams.
This value is required to calculate nutritional information accurately.
Step 6: Nutrition Calculation
Using the nutrition data stored for each category, the system calculates the total nutrients
based on the entered quantity.
This includes energy, protein, fats, carbohydrates, and other nutrients.
Step 7: Visualization
The system creates a simple pie chart to show the nutrition breakdown clearly.
This helps the user quickly understand which nutrients are present and in what amount.
Step 8: Final Output
Finally, the system displays all the details together:
• the identified food category,
• the prediction confidence,
• the calculated nutrition values, and
• the pie chart visualization.
2. Workflow Diagram
3. Sequence of Operations
1. Start the system.
2. Upload a food image.
3. Process the image and standardize its format.
4. Predict the food item and show confidence.
5. Display image and prediction to the user.
6. Ask for food quantity input.
7. Calculate nutrition based on that quantity.
8. Generate and show pie chart.
9. Display the full nutrition report.
10. End or restart for another image.
Key Features of the System
The Food Image Recognition and Nutrition Analysis System includes several helpful features
that make it easy for users to identify food items and understand their nutritional value. The
main features are:
1. Automatic Food Recognition
The system can identify the food item directly from an image.
The user only needs to upload a picture, and the system predicts the food category on its own.
2. Confidence Score Display
Along with the predicted food item, the system shows how confident it is about the result.
This helps the user understand the accuracy of the prediction.
3. Clear Image Display
The uploaded image is shown on the result page so the user can confirm that the system
recognized the correct picture.
4. Quantity-Based Nutrition Calculation
After predicting the food, the user can enter the quantity in grams.
The system automatically calculates the nutrition values—such as energy, protein, fat,
carbohydrates, and other nutrients—according to the entered amount.
5. Nutrition Pie Chart
To make the nutrition information easy to understand, the system displays a simple pie chart.
This chart visually shows the distribution of nutrients in the food.
6. Easy-to-Use Interface
The system is designed to be simple and user-friendly.
Anyone can upload an image, enter the quantity, and view the results without needing any
technical knowledge.
7. Fast Processing
The system provides results quickly.
Food recognition and nutrition calculation happen within a short time, making it convenient
for daily use.
8. Well-Organized Output
The final result includes:
• recognized food name
• confidence percentage
• nutrition table
• pie chart
Implementation Screenshots
Testing and Validation
After building the system, testing and validation were done to check how well it recognizes
food images and how accurately it calculates nutrition values. The system was tested using
images that were not used during training to make sure it works correctly in real situations.
1. Training Accuracy
During training, the system learned from a large number of food images.
From the notebook results, the system achieved around:
• Training Accuracy: 85% – 90%
This means the system correctly recognized most of the training images.
2. Testing (Validation) Accuracy
The system was also tested on a different set of images that it had never seen before.
This helps check how well the system works on new images.
From the notebook:
• Testing Accuracy: 70% – 75%
This shows that the system performs well and can correctly recognize many food
items even when the images are completely new.
3. Testing with Real Images
To check practical performance, random food pictures were uploaded manually.
These images had different lighting, angles, backgrounds, and sizes.
The system was able to:
• show the uploaded image
• predict the food item
• display the confidence score
• calculate nutrition based on quantity
• generate the nutrition pie chart
Conclusion
The Food Image Recognition and Nutrition Analysis System successfully identifies food
items from images and provides a quick nutrition report based on the quantity entered by the
user. The project shows how a computer can learn from many food images and then
recognize new images on its own. The system also calculates important nutrition values and
displays them in a simple pie chart, making it easy for users to understand the nutritional
content of the food they are eating.
Through the different stages of the project—dataset preparation, preprocessing, training,
testing, prediction, and nutrition calculation—the system proved to work smoothly and
provide reliable results. The training and testing accuracy show that the model has learned
well and can correctly identify many types of food. The testing with real images further
confirms that the system works in practical situations.
Overall, the project achieves its goal of reducing manual effort and helping users get quick
food recognition and nutrition details. It can be useful for diet tracking, health awareness, and
food-related applications. With more data and future improvements, the system can become
even more accurate and support a wider range of food categories.
Future Scope
In the future, this system can be improved and expanded in many useful ways. As technology
continues to grow, there is a lot of scope to make the food recognition and nutrition analysis
process even better. One of the main areas of future development is increasing the number of
food categories. By adding more images and more varieties of dishes, the system can become
more flexible and capable of recognizing a wider range of foods. This will help the model cover
different regions, cultures, and eating habits.
Another important area for future improvement is accuracy. By training the system with a larger
and more diverse collection of images, the recognition performance can become stronger and
more dependable. Better accuracy means the system will be able to identify food items more
correctly even when the images are taken under different lighting or angles. The nutrition part
of the system can also grow in the future. More nutrients can be added to the database so that
the system provides a more complete nutritional understanding for each food item. This will
make the system more useful for people who want to maintain a healthy diet or track their daily
food intake.
The overall working process of the system can also be made faster and smoother with further
development. With improved processing methods, the system can give results quickly and in a
more refined manner. In addition to this, the system can be expanded to understand more
complex food images, such as images that contain multiple food items or different presentation
styles. With additional work, it may also become capable of giving simple suggestions related
to nutrition or portion sizes, making it more helpful for health awareness.
In the long term, the system can become a useful tool for everyday life, helping people
understand their food better and make informed decisions. As more data becomes available and
learning methods become stronger, the system has the potential to grow into a more complete
and reliable solution for food recognition and nutrition analysis.
References
1. Bossard, L., Guillaumin, M., & Van Gool, L. Food-101 – Mining Discriminative
Components with Random Forests.
Research paper presenting the Food-101 dataset, one of the most widely used datasets
for food image classification.
2. Image Processing and Machine Learning textbooks containing theory on image
features, classification methods, and dataset preparation.
3. Educational resources explaining the steps of training, validating, and testing learning-
based image recognition systems.
4. Study materials and online tutorials related to data preprocessing, image resizing,
normalization, and augmentation techniques.
5. Standard documentation for Python scientific libraries used in machine learning, image
handling, numerical operations, and plotting.
6. Notes and articles on nutrition science that describe nutrient values, food composition,
and calculation methods used for nutritional analysis.
7. General articles and academic content discussing the role of deep learning in image
understanding and pattern recognition.