computer vision – Jeff Thompson

November 25, 2016November 25, 2016

Inpainting

Totally obsessed with the idea of inpainting, used to reconstruct and remove items from images automatically. I’m especially in love with the glitchy, Predator-esque ghosting that remains.

This is very easy to accomplish with OpenCV and a few lines of code – here I’ve run this with a radius of 5px. The results are actually pretty usable, especially in busy or small areas of the image.

The very simple code to run everything:

import numpy as np
import cv2

input_image = 'bedroom.jpg'
mask_image =  'bedroom-mask3.jpg'
radius =      5

img = cv2.imread(input_image)
mask = cv2.imread(mask_image)
mask = cv2.cvtColor(mask, cv2.COLOR_BGR2GRAY)
dest = cv2.inpaint(img, mask, radius, cv2.INPAINT_TELEA)

cv2.imwrite('Telea-' + str(radius) + '.jpg', dest)

import numpy as np

import cv2

input_image = 'bedroom.jpg'

mask_image = 'bedroom-mask3.jpg'

radius = 5

img = cv2.imread(input_image)

mask = cv2.imread(mask_image)

mask = cv2.cvtColor(mask, cv2.COLOR_BGR2GRAY)

dest = cv2.inpaint(img, mask, radius, cv2.INPAINT_TELEA)

cv2.imwrite('Telea-' + str(radius) + '.jpg', dest)

And here’s the same input, but with a really weird, blobby, random mask:

h/t Miriam Redi

August 4, 2016August 4, 2016

Facebook Image Speculation

Slow internet today reveals weird tags added by Facebook to images, guessing what they might contain. Sadly, this is not visible in the HTML source, so no good way to access it.

Edit: it appears this is a usability feature for visually-impaired users, letting them hear what might be in the image.

February 23, 2016February 23, 2018

Flickr “Magic View”

When you visit us, you can expect our auto parts for sale Indianapolis in and knowledgeable staff to be on hand. Whether you are a DIY car enthusiast or someone who is on a tight budget, we can help.

wp-image-6058″ src=”http://www.jeffreythompson.org/blog/wp-content/uploads/2016/02/FlickrMagicView_VehicleBike-web.png” alt=”FlickrMagicView_VehicleBike-web” width=”660″ height=”219″ />

Images via Flickr’s image-recognition auto-tagging system called “Magic View”.

January 12, 2015January 12, 2015

Ninja Turtles Motion Capture Suit

Somehow missed these images of the TMNT digitizing suits.

November 10, 2014

Recognizing The iPhone

<?xml version="1.0"?>
<opencv_storage>
<cascade>
  <stageType>BOOST</stageType>
  <featureType>HAAR</featureType>
  <height>32</height>
  <width>32</width>
  <stageParams>
    <boostType>GAB</boostType>
    <minHitRate>9.9500000476837158e-01</minHitRate>
    <maxFalseAlarm>5.0000000000000000e-01</maxFalseAlarm>
    <weightTrimRate>9.4999999999999996e-01</weightTrimRate>
    <maxDepth>1</maxDepth>
    <maxWeakCount>100</maxWeakCount></stageParams>
  <featureParams>
    <maxCatCount>0</maxCatCount>
    <featSize>1</featSize>
    <mode>BASIC</mode></featureParams>
  <stageNum>3</stageNum>
  <stages>
    <!-- stage 0 -->
    <_>
      <maxWeakCount>1</maxWeakCount>
      <stageThreshold>1.</stageThreshold>
      <weakClassifiers>
        <_>
          <internalNodes>
            0 -1 0 6.3478387892246246e-03</internalNodes>
          <leafValues>
            -1. 1.</leafValues></_></weakClassifiers></_>
    <!-- stage 1 -->
    <_>
      <maxWeakCount>1</maxWeakCount>
      <stageThreshold>1.</stageThreshold>
      <weakClassifiers>
        <_>
          <internalNodes>
            0 -1 2 -6.0746502876281738e-01</internalNodes>
          <leafValues>
            1. -1.</leafValues></_></weakClassifiers></_>
    <!-- stage 2 -->
    <_>
      <maxWeakCount>1</maxWeakCount>
      <stageThreshold>1.</stageThreshold>
      <weakClassifiers>
        <_>
          <internalNodes>
            0 -1 1 -2.4005537852644920e-02</internalNodes>
          <leafValues>
            1. -1.</leafValues></_></weakClassifiers></_></stages>
  <features>
    <_>
      <rects>
        <_>
          0 0 12 3 -1.</_>
        <_>
          4 0 4 3 3.</_></rects>
      <tilted>0</tilted></_>
    <_>
      <rects>
        <_>
          0 0 16 4 -1.</_>
        <_>
          0 0 8 2 2.</_>
        <_>
          8 2 8 2 2.</_></rects>
      <tilted>0</tilted></_>
    <_>
      <rects>
        <_>
          0 0 30 19 -1.</_>
        <_>
          10 0 10 19 3.</_></rects>
      <tilted>0</tilted></_></features></cascade>
</opencv_storage>

<?xml version="1.0"?>

<opencv_storage>

<stageType>BOOST</stageType>

<mode>BASIC</mode></featureParams>

<_>

<_>

0 -1 0 6.3478387892246246e-03</internalNodes>

-1. 1.</leafValues></_></weakClassifiers></_>

<_>

<_>

0 -1 2 -6.0746502876281738e-01</internalNodes>

1. -1.</leafValues></_></weakClassifiers></_>

<_>

<_>

0 -1 1 -2.4005537852644920e-02</internalNodes>

1. -1.</leafValues></_></weakClassifiers></_></stages>

<_>

<rects>

<_>

0 0 12 3 -1.</_>

<_>

4 0 4 3 3.</_></rects>

<_>

<rects>

<_>

0 0 16 4 -1.</_>

<_>

0 0 8 2 2.</_>

<_>

8 2 8 2 2.</_></rects>

<_>

<rects>

<_>

0 0 30 19 -1.</_>

<_>

10 0 10 19 3.</_></rects>

</opencv_storage>

I’m spending the month of July in Utrecht, Netherlands as artist-in-residence at Impakt, working on a project to let computers recognize other computers. For early testing, I’ve trained a computer vision “cascade file” (above) for recognizing front-facing iPhones.

The resulting cascade files, like above, will be released as a printed book, as well as on GitHub.

July 9, 2014