{"id":2347,"date":"2012-08-20T18:43:49","date_gmt":"2012-08-21T01:43:49","guid":{"rendered":"http:\/\/mcclanahoochie.com\/blog\/?post_type=portfolio&p=2347"},"modified":"2023-06-10T10:32:42","modified_gmt":"2023-06-10T17:32:42","slug":"local-contrast-enhancement-with-arrayfire-opencv","status":"publish","type":"post","link":"https:\/\/mcclanahoochie.com\/blog\/2012\/08\/local-contrast-enhancement-with-arrayfire-opencv\/","title":{"rendered":"Local Contrast Enhancement with ArrayFire + OpenCV"},"content":{"rendered":"

About one year ago, I wrote about a simple example of Image Processing with LibJacket + OpenCV<\/a>… and the trend continues today. In this post, I demonstrate how ArrayFire<\/a><\/strong> (an improved version of LibJacket) can easily interop with OpenCV, through a simple example of unsharp maksing<\/strong> (local contrast enhancement<\/a>).<\/em><\/p>\n

\u00a0<\/em><\/strong><\/h3>\n
\"\"<\/a>
Top Left: original webcam image | Top Right: histogram of V channel of input webcam image | \u00a0Bottom Left: image after LCE applied | Bottom Right: histogram of V channel of LCE image<\/figcaption><\/figure>\n

Program Flow:<\/em><\/strong><\/p>\n

    \n
  1. Capture steaming webcam video via OpenCV<\/li>\n
  2. Convert current frame OpenCV Mat to an ArrayFire array<\/a><\/em><\/li>\n
  3. Change colorspace from RGB to HSV<\/em><\/li>\n
  4. Apply unsharp masking<\/a> (LCE) to the V-channel (based on slider-bar value)<\/li>\n
  5. Convert back to RGB<\/li>\n
  6. Compute before<\/em> and after<\/em> V-channel histograms<\/li>\n
  7. Display images and plot V-channel histograms<\/li>\n
  8. Use OpenCV’s trackbar to interactively<\/em> adjust strength of LCE live<\/li>\n<\/ol>\n
    <\/div>\n
    \"\"<\/a>
    LCE adjustment value of 1.0<\/figcaption><\/figure>\n
    \"\"<\/a>
    LCE adjustment value of 5.0<\/figcaption><\/figure>\n

    The 2 screenshots above show unsharp mask adjustment values of 1.0<\/em> and 5.0<\/em> (amount of difference layer to multiply back into original). It is interesting to see how the histogram gets\u00a0progressively\u00a0smoother<\/em> as the “enhancement” factor increases.<\/p>\n

     <\/p>\n

    OpenCV Mat to Arrayfire ArraY Conversion<\/em><\/strong><\/h3>\n

    OpenCV uses a row-major, packed BGR<\/em> memory layout, where as ArrayFire uses a column-major, channel based<\/em> (RGB are separate planes) memory layout. The conversion between the two is pretty simple; here is a snippit from the code running in the demo above:<\/p>\n

    \"\"<\/a><\/p>\n

     <\/p>\n

    Try it out for yourself!<\/strong><\/em><\/h3>\n