Competition Name: The Nature Conservancy Fisheries Monitoring

Method Name:

Pre-processing: Median Filtering, Histogram Equalization
Segmentation: K-Means Clustering
Feature Extraction: Oriented FAST and Rotated BRIEF (ORB)
Classifier:K-Nearest Neighbour

Dataset Description:

The dataset compiled by The Nature Conservancy in partnership with Satlink, Archipelago Marine Research, the Pacific Community, the Solomon Islands Ministry of Fisheries and Marine Resources, the Australia Fisheries Management Authority, and the governments of New Caledonia and Palau.

The Train and Test dataset consists of images having only one fish category out of eight different categories mentioned as follows:
1. Albacore tuna
2. Bigeye tuna
3. Yellowfin tuna
4. MahiMahi
5. Opah
6. Sharks
7. Other (i.e. fish present but not the defined categories)
8. No Fish (i.e. no fish is in the image)

Method Used:

Algorithm

Input: Training Set and Test Set of Images
Output: Classification Score

1. Read the input RGB image
2. Perform median filtering for noise removal
3. Perform histogram equalization for image enhancement
4. Perform the K-Means clustering for segmentation of pre-processed image
5. Applying Oriented FAST and Rotated BRIEF (ORB) method for feature extraction
6. Apply K-Nearest Neighbour (KNN) classifier for recognition

The algorithm initiates into reading the input RGB images from the training set and store them into array. A 5X5 window-based 2D convolution operation is applied on the input images for median filtering. In this process, the neighboring pixels are ordered according to their intensity values and the median value becomes the resulting output for the central pixel of the defined window. Median filters can handle efficiently with noise, in particular impulse noise in which some individual pixels have extreme values while preserving the contrast within the images.
Contrast is a significant image element that can be defined as a ratio between the highest and the lowest pixel intensities of an image. As the images usually suffer from poor image quality, degradation in contrast and happening of shading and artefacts, the lack in centering pixel intensity, poor lightening, specimen spotting are important factors that affect the images which leads to enhance the contrast. In our methodology, histogram equalization is applied for image enhancement. Histogram of images provides a pixel-wise intensity distribution and description of the image appearance globally. The equalization process mapping histogram distribution to a wider and more uniform distribution of intensity values, as a result the intensity values are spread over the whole range of intensity distribution of the images.

The pre-processed images are segmented based on K-Means clustering approach for extraction of the region of interest. K-Means is an unsupervised learning process which assigns and groups the image pixels into a predefined number of clusters based on the calculated similarity for each pixel associate with the nearest center pixel. When all image pixels have been assigned, this iterative process continues with the new calculated center pixels. The process initiates the center pixels corresponding to the clusters randomly and keeps reassigning until criterion is met:

The K-Means objective function:

where is a chosen distance measure between a data point and the cluster centre , is an indicator of the distance of the n data points from their respective cluster centres.

After applying segmentation process, the Oriented FAST and Rotated BRIEF (ORB) method is used to extract significant features. This method combines FAST（Feature from Accelerated Segment test) keypoint detector and BRIEF (Binary Robust Independent Elementary Features) descriptor approach. This method initiates with identification of FAST corner keypoints within the image region based on comparison of the intensity threshold value between the center pixel and those in neighborhood around that pixel. Harris corner extracts the first target N number of FAST keypoints in order to employ them for further processing. A scale pyramid process is incorporated in order to extract N number of FAST keypoints at each level in the pyramid. The intensity centroid value is also measured for each corner orientation for achieving the direction of corresponding keypoints. BRIEF extracts descriptors around selected key feature points through binary coding.

Where, p(x) is the pixel intensity at that point x in image region
p(y) is the pixel intensity at that point y in image region
a set of points can uniquely identify one binary detection τ
K-nearest Neighbor algorithm incorporates the extracted key feature points in order to classify the test image data based on the trained information.

For more info, Pl do visit us www.findtutoronline.net

Also do visit our Blog : findtutoronline.net/blog

Those CQF Delegates who got Extension of CQF EXAM 1 January 2019 Cohort Solutions Guide, Pl do get in touch with us : findtutoronline.net@gmail.com or cqfquantsolutions@gmail.com

Good discount facility available for CQF Delegates who booked all CQF EXAM 1-3 & final Projects at a time. So, Hurry up to book your Order CQF Exam 1 Solutions Guide with Extension January 2019 Cohort which is readily available to delivery to your Inbox. For Instant Quotation, Give us a call or WhatsApp +91 8697669523. Instant delivery to your Inbox after funds transfer. We Accept payments through PAYPAL or Bank Transfer. (#Findtutoronline is not part of CQF nor any official support of CQF nor endorsed by CQF. #findtutoronline is an Independent International website having headquarter at North America)

WhatsApp us : +91 8697669523

Pl do visit our FACEBOOK page : https://www.facebook.com/cqfmodulesolutions

Pl do visit us our LinkedIn page : https://www.linkedin.com/in/mahesh-sarkar-mba-cqf-ca-cfa-usa-6919367b/

Keywords : Matlab Code, Python Code, CQF final Projects, CQF Help, CQF module Exam Solutions, Matlab Homework help, python homework help, matlab freelancers, matlab teachers in India, matlab tutors in India, Matlab online tutoring, matlab tutors worldwide, Matlab coding & report writing services, Advertisement Enquiry.

Pl do follow us on LinkedIn, Facebook, Twitter, Instagram

Our Instagram page : https://www.instagram.com/findtutoronline.net_2013/

Our Matlab page on Facebook : https://www.facebook.com/Matlab-at-Findtutoronlinenet-158849741591616/

Nota Bene: Academic Writing Companies Can Float their Advertisement at findtutoronline.net for Text Link Posting, Guest Post, Blog Post, Article post, Banner Post, content Writing at a reasonable Prices.

#findtutoronline Admin Team

www.findtutoronline.net

Skype ID : assignmenthelp_1

maheshsarkar5@gmail.com

ranadeb.kumar2

assignmenthelpuk

Mobile: +91 8697669523 WhatsApp.