Javascript，Tesseract，AWS IoT，Dynamodb和缓冲区

Question

Im working and testing on a small Tesseract project and with full credit to http://aalvarez.me/blog/posts/building-an-ocr-service-with-tesseractjs-in-aws-lambda.html I've got a Tesseract project running in Lambda that takes s3 events and processes the image that was uploaded to s3.

The next objective is to look at processing an image from AWS IoT. The main hurdle is that you cant send an image via MQTT.

My approach would be the send the buffer/binary stream of the image on a MQTT topic and then an IoT Rule triggers the same lambda for processing.

But 1st I needed to have a valid binary stream (no real hardware and cameras yet FYI and I'm the backend guy) to send via IoT so I modified my project to not only store the results of the OCR performed on the image but also the binary stream/buffer of the image.

I needed to do this because AWS Cloudwatch truncates the logging so when I logged out the Data payload of the s3 object that the Lambda function loads, all you get is a severely truncated buffer object.

dataBody: <Buffer 89 50 4e 47 0d 0a 1a 0a 00 00 00 0d 49 48 44 52 00 00 01 90 00 00 01 90 08 06 00 00 00 80 bf 36 cc 00 00 20 00 49 44 41 54 78 5e ed 9d 09 f8 76 df 58 ... >

So I added a bit of code that uses DynamoDB updateItem with SET as per below to set an attribute on a db item with the original buffer used to perform the OCR with as per below

            "UpdateExpression": "SET img = :attrValue",
        "ExpressionAttributeValues": {
            ":attrValue": { "B": dataBody }

That worked and the attribute was created in the item with a large string of characters which I took to be the original but in some different format perhaps

Now if I take that large string and send it as MQTT payload and trigger the rule then the Lambda fires but when I convert the payload and attempt to process it then Tesseract complains about

TypeError: Cannot read property 'mime' of null

The error is probably the result of converting the IoT payload to a Buffer for processing as its now completely different from the original Buffer when reading in the original s3 object and therefore mime property is not present. The data is now munted I'm picking

let ocrBuffer = Buffer.from(iotEvent)

ocrBuffer: <Buffer 69 56 42 4f 52 77 30 4b 47 67 6f 41 41 41 41 4e 53 55 68 45 55 67 41 41 41 5a 41 41 41 41 47 51 43 41 59 41 41 41 43 41 76 7a 62 4d 41 41 41 67 41 45 ... >

Bit of a newbie to buffers and working with images etc but I'm suspecting that the process of storing it to DynamoDB has somehow altered the original buffer or wrapped it in something so I need to unalter/unwrap it correctly so it becomes the same as the original Buffer that starts with 89 instead of 69.

Does anyone have any ideas on where its all gone wrong and where should start looking.

Right now I can upload images to s3 and I want to somehow save the payload so I can use it as a test payload for use via IoT Service

Thanks Mark

Answer 1

The crux of your problem seems to be this message from Tesseract:

TypeError: Cannot read property 'mime' of null

This at least has a clear cause. When you are sending the data to Tesseract it sounds like you are not setting the Content-type http header.

In the examples from Tesseract it looks like the Content-type header is required, which would make sense as it needs to know what kind of file you are posting to it.

So I think the question here how is your Lambda function posting data to Tesseract, and specifically how is it setting the Content-type and Content-Disposition headers?

Although there could be issues with your binary data I probably wouldn't focus on that right now, as it doesn't appear to be the thing Tesseract is complaining about.

One thing you could try is posting the (local) file to Tesseract very simply using curl. Assuming this works, compare the HTTP request with the one your Lambda function is producing and work backwards from there. I suspect you don't have a content issue, or at least that probably not the current problem.

Answer 2

Resolved

All I needed to do was specify 'base64' when converting to buffer using

Buffer.from(string, 'base64')