Source Code¶On my GitHub This code is wrapped in a class which makes it harder to post to a Notebook (a few too many lines for a post). Segmenting Fundus Images with kNN¶After preprocessing the images as described in a previous post, I tried to extract learnable features by applying the k-nearest neighbors algorithm. In… Read More »
Previous couple of posts describe some retina images pre-processing with OpenCV and IPython notebooks. Python is great but having to pre-process about 88,000 images (35,000 train and 53,000 test) I had my doubts about how long it would all take. Besides, I am a huge fan of CUDA, I have a GTX Titan GPU and… Read More »
Creating Custom Filter Banks with OpenCV¶ Suppose in order to extract curved lines from the image, we create a bank of filters (matched filters) designed to illicit a response from line segments of different orientation. To reconstruct the curves we apply each filter in the bank to the given image and then return the maximum… Read More »
Color Transfer by Histogram Specification¶Auxilary functions (described in the previous post)¶ In [1]: %matplotlib inline from matplotlib import pyplot as plt from os import path import numpy as np import cv2 import pandas as pd # display a list of images with titles def show_images(images,titles=None, scale=1.3): """Display a list of images""" n_ims = len(images) if titles… Read More »
The following few posts are a fallout from the Kaggle Diabetic Retinopathy Detection competition. I have been learning image processing with OpenCV 2.4/C++/GPU, Python 2.7, scikit-image, PIL, etc. It was a lot of fun. Finding the Eye¶We have a dataset of over 35,000 fundus images and we need to do some processing on them. Seems… Read More »
Here is a link to the winning solution for this competition. Interestingly enough, this wonderful team found that pixel intensity was a feature that worked really well in malware. This is probably why I did not completely bomb out: my “golden feature” was a 1D LBP pattern extracted from the binary images. It is quite… Read More »
This was my first competition and I finished (cue the drums) 116th/377. A spectacular failure exceeded only by the failure in my next competition. Nevertheless, here is what I did and overall I was not all that inaccurate on the validation set and probably even less inaccurate in real life, surprise surprise. This is a… Read More »
Now that I have learned how to convert IPython notebooks to WordPress-able html, here is an old post revisited. Extracting a 1D Local Binary Pattern Histogram on NVIDIA GPU with CUDA and Numbapro¶ This was done for the Microsoft Malware competition on Kaggle. In this contest, a bunch of malware files needed to be classified… Read More »