You are on page 1of 2

improvements for providing those stray animals

loving and caring homes or shelters. The images


Millions of people adopt homeless pets daily. of several animals adopted as pets and those
The visual appeal in the picture attract the who are stray animals and their metadata will be
people strongly. Picture contains a lot of analyzed to predict which images are popular.
information. We are hoping to use attractive pet Machine Learning model will be implemented on
images to save their lives. The goal is to make the datasets containing thousands of animal raw
the quality and appearance of picture better by images and the dataset will be split into a
applying different image processing techniques. balanced ratio of training and testing set. Many
A large number of stray animals are concerned Machine Learning approaches have been used
as a part of public health. Thousands of those for image processing and predictions like
animals suffer on streets and some are put to Support Vector Machines, Artificial Neural
death humanely by people. Population of stray Networks, Decision Trees, Bayesian Networks
animals is increasing drastically due to many and Regression analysis etc. We are planning
social reasons. There is a need of considering to use Convolutional Neural Networks technique
the ways to provide shelters to those stray for pet image processing. For analyzing visual
animals so as to decrease diseases in them and imagery in detail, space invariant or shift
to reduce their painful deaths in order to not only invariant Artificial neural networks which are
work for their welfare but also reducing risks of based on the shared-weight architecture are
diseases due to them in humans. Most of the used. Instead of applying classical image
people expect the animals with cute and processing, our idea is to automatically extract
attractive photos get adopted faster with more features for classifying the animals in images.
ease. This idea is defined on a method that is not
Image processing and segmentation techniques human-defined algorithm based.
are applied using different methodologies like A convolution neural network uses a
region based enhancement, colour based differentiable network that has a sequence of
enhancement, shape based enhancement, layers each with simple API. Each 3D volume
texture based enhancement and edge based from input layer is transformed to output 3D
enhancement. layer.
Image popularity is related to the estimation of
rating scores with respect to different factors.
The factors include visual content, proper
appearance, brightness, resolution and other
details. These features must be considered
wisely in the prediction. Along with these
features extraction, the selection of appropriate
model for prediction is very much crucial.
We are addressing this problem by collecting a
huge dataset containing thousands of raw
images of animals from Kaggle. Using image Fig 1: Convolution Neural Network model
processing and data science, photos of these
stray animals can help suggesting any

The model will analyze the color, dimensions features of each image will be checked as
and pixel detail of each picture to estimate inputs. Data will be stratified in a balance ratio of
image popularity scores. In CNN model, each training and testing sets. Validation phase would
input is actually a pixel of the images that are include the testing of final model.
being processed using the algorithm. All the

You might also like