{
  "id": 162470,
  "title": "Single Channel Image",
  "url": "/competitions/alaska2-image-steganalysis/discussion/162470",
  "author_name": "",
  "post_date": "2020-06-29T03:52:10.061761600Z",
  "votes": 1,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi All,</p>\n\n<p>I was wondering if single channel images would work in this competition. Do steganography algorithms do things on RGB channels so that it becomes harder to classify things with only one of the channels or grayscale images?</p>\n\n<p>If I hope to reduce memory usage, should I use a single channel or reduce image size. Which would retain most useful features?</p>",
  "messages": [
    {
      "id": "906069",
      "postDate": "06/29/2020 03:52:10",
      "content": "<p>Hi All,</p>\n\n<p>I was wondering if single channel images would work in this competition. Do steganography algorithms do things on RGB channels so that it becomes harder to classify things with only one of the channels or grayscale images?</p>\n\n<p>If I hope to reduce memory usage, should I use a single channel or reduce image size. Which would retain most useful features?</p>",
      "rawMarkdown": "Hi All,\n\nI was wondering if single channel images would work in this competition. Do steganography algorithms do things on RGB channels so that it becomes harder to classify things with only one of the channels or grayscale images?\n\nIf I hope to reduce memory usage, should I use a single channel or reduce image size. Which would retain most useful features?",
      "votes": null
    },
    {
      "id": "906622",
      "postDate": "06/29/2020 12:55:54",
      "content": "<p>I have the same question. Most of the research papers that I found so far have used single-channel images in their works. </p>",
      "rawMarkdown": "I have the same question. Most of the research papers that I found so far have used single-channel images in their works.",
      "votes": null
    },
    {
      "id": "906661",
      "postDate": "06/29/2020 13:28:22",
      "content": "<p>It would be possible to some extent if you use Y channel in YCbCr color space because most of the embedded bits seem to be in Y channel.</p>",
      "rawMarkdown": "It would be possible to some extent if you use Y channel in YCbCr color space because most of the embedded bits seem to be in Y channel.",
      "votes": null
    },
    {
      "id": "907376",
      "postDate": "06/29/2020 22:49:11",
      "content": "<p>Haven't tried single vs multi-channel but with regards to memory usage, reducing the number of image channels has pretty much no effect - you would have to reduce image/batch size to get any meaningful reduction in memory usage. However it seems to be pretty clear that resizing the image results in worse performance. </p>",
      "rawMarkdown": "Haven't tried single vs multi-channel but with regards to memory usage, reducing the number of image channels has pretty much no effect - you would have to reduce image/batch size to get any meaningful reduction in memory usage. However it seems to be pretty clear that resizing the image results in worse performance.",
      "votes": null
    },
    {
      "id": "907446",
      "postDate": "06/30/2020 01:07:34",
      "content": "<p>Are they using grayscale image or selecting a particular channel from RGB or YCbCr? <a href=\"/udaykamal\">@udaykamal</a> </p>",
      "rawMarkdown": "Are they using grayscale image or selecting a particular channel from RGB or YCbCr? @udaykamal",
      "votes": null
    },
    {
      "id": "908079",
      "postDate": "06/30/2020 11:09:57",
      "content": "<p>It seems that they were using grayscale images. However, we do have plan to try out a particular channel of YCbCr.</p>",
      "rawMarkdown": "It seems that they were using grayscale images. However, we do have plan to try out a particular channel of YCbCr.",
      "votes": null
    },
    {
      "id": "909382",
      "postDate": "06/30/2020 14:53:24",
      "content": "<p>I read somewhere in the discussion that 80 % of hidden information is in Cb, Cr channels</p>",
      "rawMarkdown": "I read somewhere in the discussion that 80 % of hidden information is in Cb, Cr channels",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 906622,
      "author_name": "udaykamal",
      "author_url": "",
      "post_date": "06/29/2020 12:55:54",
      "content": "<p>I have the same question. Most of the research papers that I found so far have used single-channel images in their works. </p>",
      "votes": null,
      "replies": [
        {
          "id": 907446,
          "author_name": "tonychenxyz",
          "author_url": "",
          "post_date": "06/30/2020 01:07:34",
          "content": "<p>Are they using grayscale image or selecting a particular channel from RGB or YCbCr? <a href=\"/udaykamal\">@udaykamal</a> </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 908079,
          "author_name": "udaykamal",
          "author_url": "",
          "post_date": "06/30/2020 11:09:57",
          "content": "<p>It seems that they were using grayscale images. However, we do have plan to try out a particular channel of YCbCr.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 906661,
      "author_name": "ren4yu",
      "author_url": "",
      "post_date": "06/29/2020 13:28:22",
      "content": "<p>It would be possible to some extent if you use Y channel in YCbCr color space because most of the embedded bits seem to be in Y channel.</p>",
      "votes": null,
      "replies": [
        {
          "id": 909382,
          "author_name": "rishabhdhiman",
          "author_url": "",
          "post_date": "06/30/2020 14:53:24",
          "content": "<p>I read somewhere in the discussion that 80 % of hidden information is in Cb, Cr channels</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 907376,
      "author_name": "vaillant",
      "author_url": "",
      "post_date": "06/29/2020 22:49:11",
      "content": "<p>Haven't tried single vs multi-channel but with regards to memory usage, reducing the number of image channels has pretty much no effect - you would have to reduce image/batch size to get any meaningful reduction in memory usage. However it seems to be pretty clear that resizing the image results in worse performance. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "906069": "Hi All,\n\nI was wondering if single channel images would work in this competition. Do steganography algorithms do things on RGB channels so that it becomes harder to classify things with only one of the channels or grayscale images?\n\nIf I hope to reduce memory usage, should I use a single channel or reduce image size. Which would retain most useful features?",
    "906622": "I have the same question. Most of the research papers that I found so far have used single-channel images in their works.",
    "906661": "It would be possible to some extent if you use Y channel in YCbCr color space because most of the embedded bits seem to be in Y channel.",
    "907376": "Haven't tried single vs multi-channel but with regards to memory usage, reducing the number of image channels has pretty much no effect - you would have to reduce image/batch size to get any meaningful reduction in memory usage. However it seems to be pretty clear that resizing the image results in worse performance.",
    "907446": "Are they using grayscale image or selecting a particular channel from RGB or YCbCr? @udaykamal",
    "908079": "It seems that they were using grayscale images. However, we do have plan to try out a particular channel of YCbCr.",
    "909382": "I read somewhere in the discussion that 80 % of hidden information is in Cb, Cr channels"
  },
  "source": "meta"
}