Questions tagged with Amazon Textract

How does textract determine when to segment text vertically or horizontally? Is there a way to help textract do this? For example, this one https://gyazo.com/358819c526664c00922c6d231ea38fe5 vs this one https://gyazo.com/0279b6e040a91392e2ccad1f6379ae31

Textract does not always work. Sometimes it fails completely. [Sometimes it fails in recognizing very obvious text](https://gyazo.com/5658baacc353349442c72f1ce5c9ac7e). [Sometimes it recognizes jibberish where there are obvious numbers](https://gyazo.com/8e3cc523552449ff54b9ed8fdbe6594f). Is it possible to get reimbursed for failed cases?

Textract results on recognizing basic arithmetic seems to degrade with color This series of images show Textract **failing unusually in all cases except the one** where the image has been both grayscale and brightness/contrast (50/50 and 25/25) - [unedited image from the camera](https://i.gyazo.com/0c6d8126dff5269dbe089a090a9e9d26.png) FAIL - [brightness contrast applied without grayscale](https://gyazo.com/8e3cc523552449ff54b9ed8fdbe6594f) FAIL - [grayscale](https://gyazo.com/8e3cc523552449ff54b9ed8fdbe6594f) FAIL - [grayscale with brightness contrast] (https://i.gyazo.com/22269131293c8e7b50c7aee7b998554c.png) finally! Is one supposed to grayscale the image before sending to textract? Should one also apply brightness/contrast? I assume Textract was trained with grayscale images - so should the service automatically convert the input images to grayscale?

Could you please add Scandinavian language support to AWS Textract service. Currently the service works quite nicely except for the specific characters å, æ, ø

I created the private team from the Amazon SageMaker console for labeling tasks followed by the creation of the human review workflow, which I later integrated with the Amazon Textract for Key-Value pair extraction. While I called the analyze_document (along with HumanLoop configuration) to extract key-value pairs for the first time it worked as expected and I was able to see the Job in the labeling project console. However, when I called it again (irrespective of the same or different file) the HumanLoop started giving the below error. "[ERROR] InvalidParameterException: An error occurred (InvalidParameterException) when calling the AnalyzeDocument operation: HumanLoop 'textractworkflow1' already exists and it is associated with a different InputContent. Please use a new HumanLoopName and try your request again." Do we have to create a new Human Review Loop each time we trigger analyze_document with another file?

Hello, I want to call textract DetectDocumentText API from Salesforce using a base64 string that represents a black png. I tried using the following http request: - Authorization headers using aws signature with secret and key - Endpoint: https://textract.us-east-2.amazonaws.com/DetectDocumentText - Method: Tried GET/POST/PUT but I am not sure which one to use. - JSON: { "Document": { "Bytes": "iVBORw0KGgoAAAANSUhEUgAAAGQAAAAKCAQAAADoFTP1AAAAIklEQVR42mNk+MkwLADjqEdGPTLqkVGPjHpk1COjHmFgAAAiwgnFqUltyAAAAABJRU5ErkJggg==" } } also tried with S3 bucket file and JSON: { "Document": { "S3Object": { "Bucket": "mybucket", "Name": "filename.jpeg" } } } Result: 200 with POST method => response: {"Output":{"__type":"com.amazon.coral.service#UnknownOperationException"},"Version":"1.0"} Apart from this, I tried calling S3 API from Salesforce and successfully uploaded a file. There, I am using the same authorization method so that seems to be fine I guess. Not sure what I am doing wrong or if something is missing in my http request. Thanks for your help. Edited by: AgustinB on Aug 31, 2021 3:12 PM Edited by: AgustinB on Aug 31, 2021 3:18 PM

I have read the instructions <https://docs.aws.amazon.com/textract/latest/dg/api-async-roles.html#api-async-roles-all-topics> . My set up is somewhat different because I am using a Cognito user. To enable textract to publish msg to SNS, I pass relevant permissions to Textract to enable it to call SNS. I am able to call StartDocumentAnalysis method and get a response. But the SNS message is never published. The weird thing is that on a few occasions I did see the several data points in CloudWatch's SNS Metric - 'NumberOfNotificationsDelivered', indicating that the messages were published. However, they are almost all gone now. What is wrong with the below? The cognito authorized user has the CognitoAuthRole role: CognitoAuthRole: Type: AWS::IAM::Role Properties: AssumeRolePolicyDocument: Version: '2012-10-17' Statement: - Effect: Allow Principal: Federated: cognito-identity.amazonaws.com Action: sts:AssumeRoleWithWebIdentity Condition: StringEquals: cognito-identity.amazonaws.com:aud: !Ref CoginitoIdentityPool ForAnyValue:StringLike: cognito-identity.amazonaws.com:amr: authenticated - Effect: Allow Principal: Service: lambda.amazonaws.com Action: sts:AssumeRole - Effect: Allow Principal: Service: textract.amazonaws.com Action: sts:AssumeRole Description: Used by cognito authenticated users ManagedPolicyArns: - !Ref DesktopPolicy #definition is immediately below And the desktop policy is: DesktopPolicy: Type: AWS::IAM::ManagedPolicy Properties: ManagedPolicyName: DesktopBackup PolicyDocument: Version: '2012-10-17' Statement: - Effect: Allow Action: - 'iam:GetRole' - 'iam:PassRole' Resource: !GetAtt "TextractEc2Role.Arn" #definition is below - Effect: Allow Action: - "sns:Publish" Resource: - arn:aws:sns:us-east-1:xxxxxxxxxxxx:AmazonTextractTopic - Effect: Allow Action: - "textract:GetDocumentAnalysis" - "textract:GetDocumentTextDetection" - "textract:StartDocumentAnalysis" - "textract:StartDocumentTextDetection" Resource: - "*" The role that is passed to Textract service using iam:PassRole is: TextractEc2Role: Type: AWS::IAM::Role Properties: AssumeRolePolicyDocument: Version: '2012-10-17' Statement: - Effect: Allow Principal: Service: textract.amazonaws.com Action: sts:AssumeRole ManagedPolicyArns: - arn:aws:iam::aws:policy/AmazonS3ReadOnlyAccess - arn:aws:iam::aws:policy/AmazonSNSFullAccess - arn:aws:iam::aws:policy/AmazonTextractFullAccess - arn:aws:iam::aws:policy/service-role/AmazonTextractServiceRole RoleName: TextractEc2 Edited by: L Jones on Sep 1, 2020 6:40 PM Edited by: L Jones on Sep 1, 2020 6:41 PM

Hi, I'm trying to use Amazon Textract to get the information from an identification card but I'm having some problem. Some information is split in the middle and read as two blocks and in some cases it doesn't see the last letter/number. Is there a way to specify/modify the bounding box to get the correct information? Or to train the AI to know where are the information for the given identification card? If not what AWS services can I use to achieve this goal beside Textract and Rekognition? If possible to have support for pdf and black and white images. Thank you for your time!

Hello, Would like to use Human review when analyzing PDFs however the current call does not have HumanLoopConfig property similar to analyze_document. Is it possible to have a HumanReview with PDFs currently? Edited by: Shipeci1 on May 6, 2020 4:10 PM Edited by: Shipeci1 on May 6, 2020 6:53 PM

Hi, I currently have a Lambda that is triggered upon uploading a PDF document to an S3 bucket. Once called, the Lambda calls the start_document_text_detection method to extract the document text. The job gets published to an SNS topic and another Lambda gets triggered on completion to get_document_text_detection method to retrieve the results and upload it into said bucket. If I upload a document to the root of the S3 bucket, everything works great. My issue is that when I create a folder within the S3 bucket and upload a PDF there, the trigger fires but I am getting the following error : ``` [ERROR] InvalidParameterException: An error occurred (InvalidParameterException) when calling the StartDocumentTextDetection operation: Request has invalid parameters Traceback (most recent call last): File "/var/task/lambda_function.py", line 33, in lambda_handler raise e File "/var/task/lambda_function.py", line 26, in lambda_handler &#39;SNSTopicArn&#39;: os.environ[&#39;sns_arn&#39;] File "/opt/python/botocore/client.py", line 357, in _api_call return self._make_api_call(operation_name, kwargs) File "/opt/python/botocore/client.py", line 661, in _make_api_call raise error_class(parsed_response, operation_name) ``` SNSTopicARN being part of the NotificationChannel argument to the start_document_text_detection call. The RoleARN also has FullSNSAccess policy and the JobTag is the document name if that helps. I checked the limits of Textract and everything is within bounds. I also tried adjusting the prefix in the Events property of the S3 bucket to match the folder, with no luck. Any advice would be appreciated.

Hi I'm trying to modify this code at https://github.com/aws-samples/amazon-textract-enhancer/blob/master/functions/detect-text-postprocess-page.py I want to be able to pipe the results of textract analysis into comprehend medical api I've tried to hacking the code, unfortunately I'm not a python expert Any ideas how best to approach this? cheers Jon from textract_util import * import io import os import json import time import boto3 def lambda_handler(event, context): #Initialize Boto Resource s3 = boto3.resource('s3') textract = boto3.client('textract') dynamodb = boto3.client('dynamodb') table_name=os.environ\['table_name'] file_list = \[] if "Records" in event: records = event\['Records'] numRecords = len(records) print("{} messages recieved".format(numRecords)) for record in records: documentBlocks = None num_pages = 0 num_lines = 0 bucket = "" upload_prefix = "" textractJobId = "" textractStatus = "" textractAPI = "" textractJobTag = "" textractS3ObjectName = "" textractS3Bucket = "" textractTimestamp = "" if 'Sns' in record.keys(): sns = record\['Sns'] if 'Message' in sns.keys(): message = json.loads(sns\['Message']) textractJobId = message\['JobId'] print("{} = {}".format("JobId", textractJobId)) textractStatus = message\['Status'] print("{} = {}".format("Status",textractStatus)) textractTimestamp = str(int(float(message\['Timestamp'])/1000)) print("{} = {}".format("Timestamp",textractTimestamp)) textractAPI = message\['API'] print("{} = {}".format("API", textractAPI)) textractJobTag = message\['JobTag'] print("{} = {}".format("JobTag", textractJobTag)) documentLocation = message\['DocumentLocation'] textractS3ObjectName = documentLocation\['S3ObjectName'] print("{} = {}".format("S3ObjectName", textractS3ObjectName)) textractS3Bucket = documentLocation\['S3Bucket'] print("{} = {}".format("S3Bucket", textractS3Bucket)) bucket = textractS3Bucket document_path = textractS3ObjectName\[:textractS3ObjectName.rfind("/")] if textractS3ObjectName.find("/") >= 0 else "" document_name = textractS3ObjectName\[textractS3ObjectName.rfind("/")+1:textractS3ObjectName.rfind(".")] if textractS3ObjectName.find("/") >= 0 else textractS3ObjectName\[:textractS3ObjectName.rfind(".")] document_type = textractS3ObjectName\[textractS3ObjectName.rfind(".")+1:].upper() if document_path == "": upload_prefix = textractJobId else: upload_prefix = "{}/{}".format(document_path, textractJobId) print("upload_prefix = " + upload_prefix) num_pages, documentBlocks = GetTextDetectionResult(textract, textractJobId) if documentBlocks is not None and len(documentBlocks) > 0: print("{} Blocks retrieved".format(len(documentBlocks))) #Extract lines of texts into a Python dictionary by parsing the raw JSON from Textract blocks = groupBlocksByType(documentBlocks) document_text, num_lines = extractTextBody(blocks) #Generate JSON document using form fields information json_document = "{}-text.json".format(document_name) json_file = open("/tmp/"_json_document,'w_') json_file.write(json.dumps(document_text, indent=4, sort_keys=True)) json_file.close() s3.meta.client.upload_file("/tmp/"+json_document, bucket, "{}/{}".format(upload_prefix,json_document)) try: response = dynamodb.update_item( TableName=table_name, Key={ 'JobId':{'S':textractJobId}, 'JobType':{'S':'TextDetection'} }, ExpressionAttributeNames={"#tf": "TextFiles", "#jst": "JobStatus", "#jct": "JobCompleteTimeStamp", "#nl": "NumLines", "#np": "NumPages"}, UpdateExpression='SET #tf = list_append(#tf, :text_files), #jst = :job_status, #jct = :job_complete, #nl = :num_lines, #np = :num_pages', ExpressionAttributeValues={ ":text_files": {"L": \[{"S": "{}/{}".format(upload_prefix,json_document)}]}, ":job_status": {"S": textractStatus}, ":job_complete": {"N": str(textractTimestamp)}, ":num_lines": {"N": str(num_lines)}, ":num_pages": {"N": str(num_pages)} } ) except Exception as e: print('DynamoDB Insertion Error is: {0}'.format(e)) else: try: response = dynamodb.update_item( TableName=table_name, Key={ 'JobId':{'S':textractJobId}, 'JobType':{'S':'TextDetection'} }, ExpressionAttributeNames={"#jst": "JobStatus", "#jct": "JobCompleteTimeStamp"}, UpdateExpression='SET #jst = :job_status, #jct = :job_complete', ExpressionAttributeValues={ ":job_status": {"S": textractStatus}, ":job_complete": {"N": str(textractTimestamp)} } ) except Exception as e: print('DynamoDB Insertion Error is: {0}'.format(e)) s3_result = s3.meta.client.list_objects_v2(Bucket=bucket, Prefix="{}/".format(upload_prefix), Delimiter = "/") if 'Contents' in s3_result: for key in s3_result\['Contents']: if key\['Key'].endswith("json"): file_list.append("https://s3.amazonaws.com/{}/{}".format(bucket, key\['Key'])) while s3_result\['IsTruncated']: continuation_key = s3_result\['NextContinuationToken'] s3_result = s3.meta.client.list_objects_v2(Bucket=bucket, Prefix="{}/".format(upload_prefix), Delimiter="/", ContinuationToken=continuation_key) for key in s3_result\['Contents']: if key\['Key'].endswith("json"): file_list.append("https://s3.amazonaws.com/{}/{}".format(bucket, key\['Key'])) print(file_list) return file_list thanks Jon