- autoclose_loader: true - runtimes: - src: "https://cdn.jsdelivr.net/pyodide/dev/full/pyodide.js" name: pyodide-dev lang: python - Pillow - opencv-python - numpy - paths: - /walicence.png Developer Portfolio

Projects

Zava

Zone Activity and Volumetric Analysis

Zava is an iOS and WatchOS Health and Fitness application developed by myself in partnership with its founder. Zava aims to optimise zone training and provide a user interface with unprecedented control and usability.

Languages: SwiftUI
Tools: Healthkit
Requirements: Apple Watch and iPhone.

(Under Development - Source code will not be provided but demonstrations will. Beta release 31/01/2024)

QR CODE GENERATOR

Generate your own free QR code by pasting a link into the text box.

Loading...

After I realised that the QR code I used on my resume had expired because I did not pay a fee, I decided to generate my own QR codes using an existing QR API. Go ahead, create one in the window, its on the house.

Languages: Javascript, Tailwind CSS
Tools: QRCode.js

Computer Vision Licence Extraction

Card Image

This is an example before and after of document alignment


#!/home/dh_tp7g6m/opt/python-3.11.4/bin/python3 from js import document, console, Uint8Array, window, File import asyncio import io import os import numpy from pyodide.ffi import create_proxy import cv2 from PIL import Image, ImageFilter from pyodide.http import pyfetch async def _upload_change_and_show(e): #Get the first file from upload file_list = e.target.files first_item = file_list.item(0) #Get the data from the files arrayBuffer as an array of unsigned bytes array_buf = Uint8Array.new(await first_item.arrayBuffer()) #BytesIO wants a bytes-like object, so convert to bytearray first bytes_list = bytearray(array_buf) my_bytes = io.BytesIO(bytes_list) #Create PIL image from np array my_image = Image.open(my_bytes) console.log(f"{my_image.format= } {my_image.width= } {my_image.height= }") my_image = numpy.array(my_image) # Convert RGB to BGR my_image = my_image[:, :, ::-1].copy() console.log("before template") #get template image. temp = await get_template() console.log("after temp return") #align my_image = await find_document_edges_and_align(my_image, temp , 0.1) #convert back to PIL console.log("After calling align in upload") color_converted = cv2.cvtColor(my_image, cv2.COLOR_BGR2RGB) my_image = Image.fromarray(color_converted) #Convert Pillow object array back into File type that createObjectURL will take my_stream = io.BytesIO() my_image.save(my_stream, format="PNG") #Create a JS File object with our data and the proper mime type image_file = File.new([Uint8Array.new(my_stream.getvalue())], "new_image_file.png", {type: "image/png"}) #Create new tag and insert into page document.getElementById("output_upload_pillow").innerHTML = "" new_image = document.createElement('img') new_image.src = window.URL.createObjectURL(image_file) document.getElementById("output_upload_pillow").appendChild(new_image) # Run image processing code above whenever file is uploaded upload_file = create_proxy(_upload_change_and_show) document.getElementById("file-upload-pillow").addEventListener("change", upload_file) async def get_template(): temp = numpy.array(Image.open('walicence.png')) return temp #This function aligns and crops the image async def find_document_edges_and_align(cv_image, template, keepPercent): # Read the input image maxFeatures = 4000 img = cv_image console.log("ABove gray conversion in align") # Convert the image to grayscale gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) templateGray = cv2.cvtColor(template, cv2.COLOR_BGR2GRAY) console.log("Below gray conversion in align") # Apply GaussianBlur to reduce noise and help edge detection blurred = cv2.GaussianBlur(gray, (3, 3), 0 ) temp_blur = cv2.GaussianBlur(templateGray, (3,3), 0) # Detetct orb features orb = cv2.ORB_create(maxFeatures) (kpsA, descsA) = orb.detectAndCompute(blurred, None) (kpsB, descsB) = orb.detectAndCompute(temp_blur, None) #match features method = cv2.DESCRIPTOR_MATCHER_BRUTEFORCE_HAMMING matcher = cv2.DescriptorMatcher_create(method) matches = matcher.match(descsA, descsB, None) # sort the matches by their distance (the smaller the distance, # the "more similar" the features are) matches = sorted(matches, key=lambda x: x.distance) # keep only the top matches keep = int(len(matches) * keepPercent) matches = matches[:keep] # check to see if we should visualize the matched keypoints matchedVis = cv2.drawMatches(img, kpsA, template, kpsB, matches, None) # allocate memory for the keypoints (x, y)-coordinates from the # top matches -- we'll use these coordinates to compute our # homography matrix ptsA = numpy.zeros((len(matches), 2), dtype="float") ptsB = numpy.zeros((len(matches), 2), dtype="float") # loop over the top matches for (i, m) in enumerate(matches): # indicate that the two keypoints in the respective images # map to each other ptsA[i] = kpsA[m.queryIdx].pt ptsB[i] = kpsB[m.trainIdx].pt # compute the homography matrix between the two sets of matched # points (H, mask) = cv2.findHomography(ptsA, ptsB, method=cv2.RANSAC) # use the homography matrix to align the images (h, w) = template.shape[:2] aligned = cv2.warpPerspective(img, H, (w, h)) # return the aligned image console.log("returning aligned") return aligned

Drivers Licence Text Extraction

This model utilises pyscript to execute all computer vision tasks inside the browser as opposed to server side.
Submit an image of a West Australian Driver's Licence and the web app will align and crop the licence. I suggest googling an image of a sample licence. Currently the alignment model is sensitive so there is a threshold to the quality of images it can successfuly align and crop. Having said that, most automatic document alignment/extraction services require some degree of quality control over the submissions.

Breakdown: using orb feature detection with a template licence to perpective transform your image and crop it.
Extraction still to be implemented.
Pre-measured bounding boxes from my template licence will match the key-value pairs.(coming soon)


Languages: Python
Tools: Open CV, PIL

Source Code

This Portfolio Website!

This website was built from scratch and deployed through Wordpress. I had previously used a Wordpress template to build my band website but decided this was a good reason to build one from scratch.


Languages: Html, CSS, Javascript, python