Scribd
Upload
29 views15 pages

Project

The document discusses image classification and leaf species recognition using computer vision and machine learning techniques. It describes image classification as a supervised learning problem where a model is trained on labeled image data to learn how to classify new images. It then discusses building a system to classify plant leaf images and use the predictions to provide farmers and gardeners information to increase productivity. Key steps involved are extracting features from leaf images like texture, shape, and combining different feature descriptors to effectively represent the images for classification.

Uploaded by

debojyoti
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
29 views15 pages

Project

The document discusses image classification and leaf species recognition using computer vision and machine learning techniques. It describes image classification as a supervised learning problem where a model is trained on labeled image data to learn how to classify new images. It then discusses building a system to classify plant leaf images and use the predictions to provide farmers and gardeners information to increase productivity. Key steps involved are extracting features from leaf images like texture, shape, and combining different feature descriptors to effectively represent the images for classification.

Uploaded by

debojyoti
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 15

HEADING/TITLE

• What is Image Classification?


• The ability of a machine learning model
to classify or label an image into its
respective class with the help of learned
features from hundreds of images is
called as Image Classification.
• This is a supervised learning problem
where we humans must provide training
data (set of images along with its labels)
to the machine learning model so that it
learns how to discriminate each image
(by learning the pattern behind each
image) with respect to its label.
LEAF SPECIES RECOGNITION
• Project Idea
To build an intelligent system that was trained with massive dataset
of plant leaf images.
• The system predicts the label/class of the plant using Computer
Vision techniques and Machine Learning algorithms.
• The system searches the web for all the plant related data after
predicting the label/class of the captured image.
• Your system helps gardeners and farmers to increase their
productivity and yield with the help of automating tasks in
garden/farm.
• Your system applies the recent technological advancements such as
Internet of Things (IoT) and Machine Learning in the agricultural
domain.
• You build such a system for your home or your garden to monitor
your plants using a Raspberry Pi.
CLASSIFICATION PROBLEM
• Plant or Flower Species Classification is one of the
most challenging and difficult problems in
Computer Vision due to a variety of reasons.
1)Availability of plant dataset
Collecting plant dataset is a time-consuming task.
2)Millions of plant species around the world
To accomodate every such species, we need to
train our model with such large number of
images with its labels which needs tremendous
computing power.
3)High inter-class as well as intra-class variation

4) Fine-grained classification problem

5) Segmentation, View-point, Occlusion,


Illumination etc.
Segmenting the leaf region from an image is a
challenging task. This is because we might
need to remove the unwanted background
and take only the foreground object (leaf)
which is again a difficult thing due to the
shape of leaf.
FEATURE EXTRACTION

1) Features are the information or list of


numbers that are extracted from an image.
These are real-valued numbers (integers, float
or binary).
2) When deciding about the features that could
quantify a leaf, we could possibly think
of Texture and Shape as the primary ones.
3)But this approach is less likely
to produce good results, if we
choose only one feature vector.
So, we need to quantify the
image by combining different
feature descriptors so that it
describes the image more
effectively.
GLOBAL FEATURE DESCRIPTORS

• These are the feature descriptors that quantifies


an image globally. It takes in the entire image for
processing. Some of the commonly used global
feature descriptors are:-
1)Shape - Hu Moments, Zernike Moments
2)Texture - Haralick Texture, Local Binary
Patterns(LBP)
3)Others - Histogram of Oriented Gradients (HOG),
Threshold Adjancency Statistics (TAS)
LOCAL FEATURE DESCRIPTORS
• These are the feature descriptors that quantifies
local regions of an image. Interest points are
determined in the entire image and image
patches/regions surrounding those interest points
are considered for analysis. Some of the commonly
used local feature descriptors are:-

1)SIFT (Scale Invariant Feature Transform)

2)SURF(Speeded Up Robust Features)

3)ORB(Oriented Fast and Rotated BRIEF)

4)BRIEF(Binary Robust Independed Elementary Features)


COMBINING GLOBAL FEATURES
• There are two popular ways to combine these feature
vectors.
• For global feature vectors, we just concatenate each
feature vector to form a single global feature vector. This is
the approach we will be using in this tutorial.
• For local feature vectors as well as combination of global
and local feature vectors, we need something called as Bag
of Visual Words (BOVW). This approach is not discussed in
this tutorial, but there are lots of resources to learn this
technique. Normally, it uses Vocabulary builder, K-Means
clustering, Linear SVM, and Td-Idf vectorization.
GLOBAL FEATURE EXTRACTION
We will use a simpler approach to produce a baseline accuracy
for our problem. It means everything should work somehow
without any error. The folder structure for this example is
given below. Create the folders and files as shown.
• Line 1 - 8 imports the necessary libraries we need
to work with.
• Line 11 used to convert the input image to a fixed
size of (500, 500).
• Line 14 is the path to our training dataset.
• Line 17 is the number of trees we need to initialize
for Random Forests classifier.
• Line 20 is the number of bins for color histograms.
• Line 23 is the amount of training data and testing
data split (0.10 means splitting the training dataset
as 90% train data and 10% test data).
• Line 26 is the seed we keep to reproduce same
results everytime we run this script.
FUNCTIONS FOR GLOBAL FEATURE DESCRIPTORS
1) HU MOMENTS
> To extract Hu Moments features from the image, we
use cv2.HuMoments()
function provided by OpenCV.
> The argument to this function is the moments of the
image cv2.moments()
flatenned. It means we compute the moments of the image
and convert it to a vector using flatten().
> Before doing that, we convert our color image into a grayscale
image as moments expect images to be grayscale.
2)HARALICK TEXTURES
• To extract Haralick Texture features from the image, we
make use of mahotas library. The function we will be
using is mahotas.features.haralick(). Before doing that,
we convert our color image into a grayscale image as
haralick feature descriptor expect images to be grayscale.

You might also like