Description
The processing of documents for the purpose of discovering knowledge from them in an automated fashion is a challenging task and hence an open issue for the research community. Sometimes, the quality of the input images makes it much more difficult to perform these procedures correctly.
To avoid this, it is proposed the use of different image processing algorithms on documents images to improve its quality and the performance of the next step of computer vision algorithms as text detection, text recognition, ocr, table detection… Some of these image processing algorithms included in this project are: scale image, adaptive thresholding, erosion, dilation, remove objects, median blur and gpu image transformation.
Predicted Entities
How to use
binary_to_image = BinaryToImage() \
.setInputCol("content") \
.setOutputCol("image")
scaled_image_df = ImageTransformer() \
.addScalingTransform(2) \
.setInputCol("image") \
.setOutputCol("scaled_image") \
.setImageType(ImageType.TYPE_BYTE_GRAY) \
.transform(image_df)
thresholded_image = ImageTransformer() \
.addAdaptiveThreshold(21, 20)\
.setInputCol("image") \
.setOutputCol("thresholded_image") \
.transform(image_df)
eroded_image = ImageTransformer() \
.addErodeTransform(2,2)\
.setInputCol("image") \
.setOutputCol("eroded_image") \
.transform(image_df)
dilated_image = ImageTransformer() \
.addDilateTransform(1, 2)\
.setInputCol("image") \
.setOutputCol("dilated_image") \
.transform(image_df)
removebg_image = ImageTransformer() \
.addScalingTransform(2) \
.addAdaptiveThreshold(31, 2)\
.addRemoveObjects(10, 500) \
.setInputCol("image") \
.setOutputCol("corrected_image") \
.transform(image_df)
deblured_image = ImageTransformer() \
.addScalingTransform(2) \
.addMedianBlur(3) \
.setInputCol("image") \
.setOutputCol("corrected_image") \
.transform(image_df)
multiple_image = GPUImageTransformer() \
.addScalingTransform(8) \
.addOtsuTransform() \
.addErodeTransform(3, 3) \
.setInputCol("image") \
.setOutputCol("multiple_image") \
.transform(image_df)
pipeline_scaled = PipelineModel(stages=[
binary_to_image,
scaled_image_df
])
pipeline_thresholded = PipelineModel(stages=[
binary_to_image,
thresholded_image
])
pipeline_eroded = PipelineModel(stages=[
binary_to_image,
eroded_image
])
pipeline_dilated = PipelineModel(stages=[
binary_to_image,
dilated_image
])
pipeline_removebg = PipelineModel(stages=[
binary_to_image,
removebg_image
])
pipeline_deblured = PipelineModel(stages=[
binary_to_image,
deblured_image
])
pipeline_multiple = PipelineModel(stages=[
binary_to_image,
multiple_image
])
image_path = pkg_resources.resource_filename("sparkocr", "resources/ocr/images/check.jpg")
image_example_df = spark.read.format("binaryFile").load(image_path)
result_scaled = pipeline_scaled.transform(image_example_df).cache()
result_thresholded = pipeline_thresholded.transform(image_example_df).cache()
result_eroded = pipeline_eroded.transform(image_example_df).cache()
result_dilated = pipeline_dilated.transform(image_example_df).cache()
result_removebg = pipeline_removebg.transform(image_example_df).cache()
result_deblured = pipeline_deblured.transform(image_example_df).cache()
result_multiple = pipeline_multiple.transform(image_example_df).cache()
val binary_to_image = new BinaryToImage()
.setInputCol("content")
.setOutputCol("image")
val scaled_image_df = new ImageTransformer()
.addScalingTransform(2)
.setInputCol("image")
.setOutputCol("scaled_image")
.setImageType(ImageType.TYPE_BYTE_GRAY)
.transform(image_df)
val thresholded_image = new ImageTransformer()
.addAdaptiveThreshold(21, 20)
.setInputCol("image")
.setOutputCol("thresholded_image")
.transform(image_df)
val eroded_image = new ImageTransformer()
.addErodeTransform(2,2)
.setInputCol("image")
.setOutputCol("eroded_image")
.transform(image_df)
val dilated_image = new ImageTransformer()
.addDilateTransform(1, 2)
.setInputCol("image")
.setOutputCol("dilated_image")
.transform(image_df)
val removebg_image = new ImageTransformer()
.addScalingTransform(2)
.addAdaptiveThreshold(31, 2)
.addRemoveObjects(10, 500)
.setInputCol("image")
.setOutputCol("corrected_image")
.transform(image_df)
val deblured_image = new ImageTransformer()
.addScalingTransform(2)
.addMedianBlur(3)
.setInputCol("image")
.setOutputCol("corrected_image")
.transform(image_df)
val multiple_image = new GPUImageTransformer()
.addScalingTransform(8)
.addOtsuTransform()
.addErodeTransform(3, 3)
.setInputCol("image")
.setOutputCol("multiple_image")
.transform(image_df)
val pipeline_scaled = new PipelineModel().setStages(Array(
binary_to_image,
scaled_image_df))
val pipeline_thresholded = new PipelineModel().setStages(Array(
binary_to_image,
thresholded_image))
val pipeline_eroded = new PipelineModel().setStages(Array(
binary_to_image,
eroded_image))
val pipeline_dilated = new PipelineModel().setStages(Array(
binary_to_image,
dilated_image))
val pipeline_removebg = new PipelineModel().setStages(Array(
binary_to_image,
removebg_image))
val pipeline_deblured = new PipelineModel().setStages(Array(
binary_to_image,
deblured_image))
val pipeline_multiple = new PipelineModel().setStages(Array(
binary_to_image,
multiple_image))
val image_path = pkg_resources.resource_filename("sparkocr", "resources/ocr/images/check.jpg")
val image_example_df = spark.read.format("binaryFile").load(image_path)
val result_scaled = pipeline_scaled.transform(image_example_df).cache()
val result_thresholded = pipeline_thresholded.transform(image_example_df).cache()
val result_eroded = pipeline_eroded.transform(image_example_df).cache()
val result_dilated = pipeline_dilated.transform(image_example_df).cache()
val result_removebg = pipeline_removebg.transform(image_example_df).cache()
val result_deblured = pipeline_deblured.transform(image_example_df).cache()
val result_multiple = pipeline_multiple.transform(image_example_df).cache()
Example
Model Information
| Model Name: | image_processing |
| Compatibility: | Healthcare NLP 4.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Language: | en |
PREVIOUSLegal Roles NER