Table extraction from images in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Table Structure Recognition

Which of the following best describes the main goal of table structure recognition in table extraction from images?

A Detecting the exact pixel coordinates of every cell boundary in the table image

B Identifying the logical rows and columns and their relationships within the table

C Converting the entire table image into a single text string without structure

D Removing all non-table elements from the image before processing

💡 Hint

Think about what 'structure' means in a table context.

Table structure recognition focuses on understanding how cells are arranged in rows and columns, which is essential for reconstructing the table logically.

❓ Predict Output

intermediate

Output of OCR on Table Image

Given the following OCR output from a table image, what is the value of cells[1][2]?

Computer Vision

cells = [['ID', 'Name', 'Age'], ['1', 'Alice', '30'], ['2', 'Bob', '25']]

A "Alice"

B "2"

C "30"

D "Bob"

💡 Hint

Remember that indexing starts at 0 and cells[row][column].

The first index selects the row, the second the column. cells[1] is the second row, and cells[1][2] is the third element in that row, which is "30".

❓ Model Choice

advanced

Choosing a Model for Table Detection in Images

You want to detect tables in scanned document images before extracting their content. Which model type is most suitable for this task?

A A convolutional neural network (CNN) trained for object detection like Faster R-CNN

B A recurrent neural network (RNN) trained for language modeling

C A generative adversarial network (GAN) for image style transfer

D A clustering algorithm like K-means on pixel intensities

💡 Hint

Think about models designed to find objects in images.

Faster R-CNN is a CNN-based object detection model that can locate tables as objects in images, making it suitable for table detection.

❓ Metrics

advanced

Evaluating Table Extraction Accuracy

Which metric is most appropriate to evaluate how well a table extraction model correctly identifies the boundaries of table cells?

A Perplexity

B BLEU score

C Mean Squared Error (MSE)

D Intersection over Union (IoU)

💡 Hint

Consider metrics used for object detection and segmentation.

IoU measures the overlap between predicted and true bounding boxes, making it ideal for evaluating cell boundary detection.

🔧 Debug

expert

Debugging Table Extraction Code Output

Consider this Python snippet that extracts text from detected table cells using OCR. What error will it raise?

Computer Vision

import pytesseract
from PIL import Image

image = Image.open('table.png')
cells = detect_cells(image)  # returns list of bounding boxes
texts = []
for cell in cells:
    crop = image.crop(cell)
    text = pytesseract.image_to_string(crop)
    texts.append(text.strip())

print(texts[10])

A IndexError because texts may have fewer than 11 elements

B TypeError because image.crop expects coordinates as a tuple, not a list

C FileNotFoundError because 'table.png' does not exist

D AttributeError because pytesseract has no attribute 'image_to_string'

💡 Hint

Check the length of texts before accessing index 10.

If cells has fewer than 11 bounding boxes, accessing texts[10] causes an IndexError.