{
  "id": 148310,
  "title": "Image Enhancement Techniques?",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/148310",
  "author_name": "Debanga Raj Neog",
  "post_date": "2020-05-03T21:42:56.105000",
  "votes": 15,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi all,</p>\n\n<p>I'm curious about what kind of image enhancement techniques you are using that have improved results!</p>\n\n<p>I would like to share one that has helped me. </p>\n\n<h2>Enhancing Radboud Samples</h2>\n\n<p>It is <code>contrast enhancement</code> using <code>img_enhanced = cv2.addWeighted( img, contrast, img, 0, brightness)</code>. It particularly helps in <code>Radboud</code> samples. I have compiled an illustrative notebook here: <a href=\"https://www.kaggle.com/debanga/let-s-enhance-the-images\">https://www.kaggle.com/debanga/let-s-enhance-the-images</a>.</p>\n\n<p>Here is a before and after result.</p>\n\n<p><img src=\"https://raw.githubusercontent.com/debanga/depurr/master/images/enhance.png\" alt=\"\"></p>\n\n<h2>Other observations</h2>\n\n<p>I also have shown in the notebook that other filters such as <code>unsharp masking</code> can be used. It enhances the features. </p>\n\n<p>Here is a before and after result.</p>\n\n<p><img src=\"https://raw.githubusercontent.com/debanga/depurr/master/images/unsharp.png\" alt=\"\"></p>\n\n<p>Also, it is worth mentioning, I attempted Histogram equilibrium but did not work.</p>\n\n<p>I will keep exploring this area! Thanks!</p>",
  "messages": [
    {
      "id": 832114,
      "postDate": "2020-05-03T21:42:56.107Z",
      "content": "<p>Hi all,</p>\n\n<p>I'm curious about what kind of image enhancement techniques you are using that have improved results!</p>\n\n<p>I would like to share one that has helped me. </p>\n\n<h2>Enhancing Radboud Samples</h2>\n\n<p>It is <code>contrast enhancement</code> using <code>img_enhanced = cv2.addWeighted( img, contrast, img, 0, brightness)</code>. It particularly helps in <code>Radboud</code> samples. I have compiled an illustrative notebook here: <a href=\"https://www.kaggle.com/debanga/let-s-enhance-the-images\">https://www.kaggle.com/debanga/let-s-enhance-the-images</a>.</p>\n\n<p>Here is a before and after result.</p>\n\n<p><img src=\"https://raw.githubusercontent.com/debanga/depurr/master/images/enhance.png\" alt=\"\"></p>\n\n<h2>Other observations</h2>\n\n<p>I also have shown in the notebook that other filters such as <code>unsharp masking</code> can be used. It enhances the features. </p>\n\n<p>Here is a before and after result.</p>\n\n<p><img src=\"https://raw.githubusercontent.com/debanga/depurr/master/images/unsharp.png\" alt=\"\"></p>\n\n<p>Also, it is worth mentioning, I attempted Histogram equilibrium but did not work.</p>\n\n<p>I will keep exploring this area! Thanks!</p>",
      "rawMarkdown": "Hi all,\n\nI'm curious about what kind of image enhancement techniques you are using that have improved results!\n\nI would like to share one that has helped me. \n\n## Enhancing Radboud Samples\n\nIt is ```contrast enhancement``` using ```img_enhanced = cv2.addWeighted( img, contrast, img, 0, brightness)```. It particularly helps in ```Radboud``` samples. I have compiled an illustrative notebook here: https://www.kaggle.com/debanga/let-s-enhance-the-images.\n\nHere is a before and after result.\n\n![](https://raw.githubusercontent.com/debanga/depurr/master/images/enhance.png)\n\n## Other observations\nI also have shown in the notebook that other filters such as ```unsharp masking``` can be used. It enhances the features. \n\nHere is a before and after result.\n\n![](https://raw.githubusercontent.com/debanga/depurr/master/images/unsharp.png)\n\nAlso, it is worth mentioning, I attempted Histogram equilibrium but did not work.\n\nI will keep exploring this area! Thanks!",
      "votes": 15
    },
    {
      "id": 835117,
      "postDate": "2020-05-06T04:15:23.007Z",
      "content": "<p>Did u try to perform a direct comparison of the model performance trained on the original and enhanced data?</p>",
      "rawMarkdown": "Did u try to perform a direct comparison of the model performance trained on the original and enhanced data?",
      "votes": 1,
      "replies": [
        {
          "id": 835203,
          "postDate": "2020-05-06T05:40:39.010Z",
          "content": "<p>Not yet! Work in progress :D</p>",
          "rawMarkdown": "Not yet! Work in progress :D",
          "votes": 2
        },
        {
          "id": 849382,
          "postDate": "2020-05-15T17:51:08.467Z",
          "content": "<p><a href=\"/debanga\">@debanga</a>, hi, any progress?</p>",
          "rawMarkdown": "@debanga, hi, any progress?\n"
        }
      ]
    },
    {
      "id": 832466,
      "postDate": "2020-05-04T06:49:05.680Z",
      "content": "<p>Do you mean Histogram Equalization?</p>",
      "rawMarkdown": "Do you mean Histogram Equalization?",
      "votes": 1,
      "replies": [
        {
          "id": 832479,
          "postDate": "2020-05-04T06:59:45.653Z",
          "content": "<p><a href=\"/venky2506\">@venky2506</a>, I tried Histogram Equalization, does not work on the data! Tried channel-wise. Let me know if there are better ways of doing histogram equalization...</p>",
          "rawMarkdown": "@venky2506, I tried Histogram Equalization, does not work on the data! Tried channel-wise. Let me know if there are better ways of doing histogram equalization...",
          "votes": 1
        }
      ]
    },
    {
      "id": 832118,
      "postDate": "2020-05-03T21:52:12.980Z",
      "content": "<p>Great</p>",
      "rawMarkdown": "Great",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 835117,
      "author_name": "Iafoss",
      "author_url": "",
      "post_date": "2020-05-06T04:15:23.007000",
      "content": "<p>Did u try to perform a direct comparison of the model performance trained on the original and enhanced data?</p>",
      "votes": 1,
      "replies": [
        {
          "id": 835203,
          "author_name": "Debanga Raj Neog",
          "author_url": "",
          "post_date": "2020-05-06T05:40:39.010000",
          "content": "<p>Not yet! Work in progress :D</p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 849382,
          "author_name": "wayfarer",
          "author_url": "",
          "post_date": "2020-05-15T17:51:08.467000",
          "content": "<p><a href=\"/debanga\">@debanga</a>, hi, any progress?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 832466,
      "author_name": "Son of Anton v3.0",
      "author_url": "",
      "post_date": "2020-05-04T06:49:05.680000",
      "content": "<p>Do you mean Histogram Equalization?</p>",
      "votes": 1,
      "replies": [
        {
          "id": 832479,
          "author_name": "Debanga Raj Neog",
          "author_url": "",
          "post_date": "2020-05-04T06:59:45.653000",
          "content": "<p><a href=\"/venky2506\">@venky2506</a>, I tried Histogram Equalization, does not work on the data! Tried channel-wise. Let me know if there are better ways of doing histogram equalization...</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 832118,
      "author_name": "Deoth GUEI",
      "author_url": "",
      "post_date": "2020-05-03T21:52:12.980000",
      "content": "<p>Great</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "832114": "Hi all,\n\nI'm curious about what kind of image enhancement techniques you are using that have improved results!\n\nI would like to share one that has helped me. \n\n## Enhancing Radboud Samples\n\nIt is ```contrast enhancement``` using ```img_enhanced = cv2.addWeighted( img, contrast, img, 0, brightness)```. It particularly helps in ```Radboud``` samples. I have compiled an illustrative notebook here: https://www.kaggle.com/debanga/let-s-enhance-the-images.\n\nHere is a before and after result.\n\n![](https://raw.githubusercontent.com/debanga/depurr/master/images/enhance.png)\n\n## Other observations\nI also have shown in the notebook that other filters such as ```unsharp masking``` can be used. It enhances the features. \n\nHere is a before and after result.\n\n![](https://raw.githubusercontent.com/debanga/depurr/master/images/unsharp.png)\n\nAlso, it is worth mentioning, I attempted Histogram equilibrium but did not work.\n\nI will keep exploring this area! Thanks!",
    "835117": "Did u try to perform a direct comparison of the model performance trained on the original and enhanced data?",
    "832466": "Do you mean Histogram Equalization?",
    "832118": "Great"
  }
}