Top 55 OpenCV Interview Questions & Answers (Python Guide)

Top 55 OpenCV Interview Questions & Answers [2024 Ultimate Guide]

Your one-stop resource to ace any Computer Vision interview.

Welcome to the definitive guide for mastering OpenCV! Whether you're a student stepping into the exciting world of computer vision, a developer preparing for a high-stakes technical interview, or a professional looking to sharpen your skills, you've landed in the right place. This post breaks down the top 55 most frequently asked OpenCV questions into easy-to-digest answers, complete with interactive code, diagrams, and best practices.

Table of Contents

1. What is OpenCV?
2. Method to read an image?
3. Integrate OpenCV with Android?
4. Best language for OpenCV?
5. Function to draw a line?
6. What is CUDA?
7. How to resize an image?
8. Why is Convolution important?
9. Detect vertical/horizontal edges?
10. Displaying image channels?
11. What is Transfer Learning?
12. Transfer Learning algorithms?
13. Explain the VGG16 model.
14. How to finetune a model?
15. Why freeze layers?
16. What is a residual network?
17. Evaluate model from graphs?
18. What is pickle?
19. What are callbacks?
20. Method to show an image?
21. What is model checkpointing?
22. How to save images?
23. Understand image shape?
24. How to process video?
25. What is image normalization?
26. Algorithm for edge detection?
27. What is OCR?
28. How to extract text data?

29. Image blurring techniques?
30. Use of Sobel operation?
31. What is a translation matrix?
32. Rotate and translate images?
33. Convert color spaces?
34. Convert to grayscale?
35. Difference between 2D & 3D images?
36. What is FPS in video?
37. How to crop an Image?
38. How many algorithms in OpenCV?
39. Types of image filters?
40. What are Haar Cascades?
41. Detect faces in images/video?
42. Recognize unique faces?
43. What is an image histogram?
44. Histogram as an image feature?
45. What is an Image Search Engine?
46. Methods to draw shapes?
47. How to increase image quality?
48. What is aspect ratio?
49. What is an Image Transform?
50. What is image zooming?
51. Why resize/skip frames for detection?
52. What is Object Detection?
53. Define Resolution.
54. Image Translation and Scaling?
55. What is image shrinking?

Part 1: OpenCV Fundamentals

1. Explain what is OpenCV?

OpenCV stands for Open Source Computer Vision Library. It is a massive, cross-platform library of programming functions aimed at real-time computer vision. In simple terms, it provides a comprehensive toolkit of algorithms to help computers "see" and interpret visual data from the world, like images and videos. It is a cornerstone technology for applications in AI, robotics, augmented reality, and more.

2. Which method of opencv is used to read an image?

The cv2.imread() function is used to read an image from a file into a NumPy array. If the image cannot be read (e.g., file not found, corrupt file), it returns `None`, so it's crucial to check the return value.

import cv2
# Read the image in color (default mode)
image = cv2.imread("path/to/your/image.jpg")

if image is not None:
    print("Image loaded successfully with shape:", image.shape)
else:
    print("Error: Could not read the image.")

3. How can you integrate OpenCV with Android?

Integrating OpenCV into an Android project involves importing the OpenCV Android SDK as a module in Android Studio. This enables powerful computer vision capabilities in mobile apps.

Key Steps:

1. Download the OpenCV Android SDK from the official OpenCV website.
2. In Android Studio, go to File → New → Import Module... and select the `sdk` directory from the unzipped package.
3. Add the newly imported module as a dependency to your app's `build.gradle` file.
4. Initialize the OpenCV library in your app's Java/Kotlin code, typically using OpenCVLoader.initDebug();.

4. Which language is best for OpenCV?

The "best" language depends entirely on the project's goals, balancing performance against development speed.

• C++: Best for performance. Since OpenCV is written in C++, using it in C++ gives maximum speed and is the top choice for production-level, real-time, or resource-constrained applications.
• Python: Best for rapid development and prototyping. Python's simple syntax and vast data science ecosystem make it ideal for research, learning, and building complex AI pipelines quickly.
• Java: The primary choice for Android development and integration into large-scale Java-based enterprise systems.

5. In OpenCV which function is used to draw a line?

The cv2.line() function is used to draw a line segment on an image. It modifies the image in-place.

# cv2.line(image, start_point, end_point, color, thickness)
# Example: Draw a blue line from (0,0) to (200,300)
# Remember: color is in BGR format (Blue, Green, Red)
cv2.line(image, (0, 0), (200, 300), (255, 0, 0), 5)

6. What is CUDA?

CUDA (Compute Unified Device Architecture) is a parallel computing platform and programming model created by NVIDIA. It allows software to utilize the massive processing power of an NVIDIA Graphics Processing Unit (GPU) for general-purpose computing. When OpenCV is compiled with CUDA support, it can offload computationally expensive operations to the GPU, leading to a huge performance boost for real-time video processing.

7. How can you do image resize in OpenCV?

Image resizing is done using the cv2.resize() function. You can resize to an absolute dimension or by a scaling factor.

Interpolation matters: For shrinking an image, cv2.INTER_AREA is recommended. For enlarging, cv2.INTER_CUBIC (slower, higher quality) or cv2.INTER_LINEAR (faster, good quality) are best.

import cv2
image = cv2.imread('image.jpg')

# Method 1: Resize to specific dimensions (width, height)
resized_abs = cv2.resize(image, (400, 300), interpolation=cv2.INTER_LINEAR)

# Method 2: Resize by a scaling factor (e.g., 50% of original)
resized_scaled = cv2.resize(image, None, fx=0.5, fy=0.5, interpolation=cv2.INTER_AREA)

8. Why is Convolution important to extract image features?

Convolution is the fundamental operation in Convolutional Neural Networks (CNNs) and is crucial for feature extraction because it allows a model to learn a hierarchy of spatial patterns automatically. It works by sliding a small filter (kernel) over the image to detect localized features like edges, corners, and textures. Deeper layers combine these simple features to recognize more complex patterns, like parts of objects and eventually entire objects.

9. How to detect vertical and horizontal edges in images/video?

Edges can be detected using the Sobel operator via the cv2.Sobel() function. By specifying the direction of the derivative (`dx` for vertical, `dy` for horizontal), you can find edges along that axis.

import cv2
image = cv2.imread('image.jpg', cv2.IMREAD_GRAYSCALE)
# Detect horizontal edges (derivative in y-direction)
sobel_y = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=5)
# Detect vertical edges (derivative in x-direction)
sobel_x = cv2.Sobel(image, cv2.CV_64F, 1, 0, ksize=5)

10. If you display each channel from a color image, which color will you see and why?

You will see three separate grayscale images, not colored images.

Why? When you split a 3D color image array into its individual channels (e.g., `b, g, r = cv2.split(image)`), each channel is a 2D array. Image display functions like `cv2.imshow()` interpret any 2D array as a grayscale image, where pixel values represent intensity (0=black, 255=white). The "Blue" channel image will simply appear brightest where the original image had the most intense blue color.

Part 2: Deep Learning & Transfer Learning

11. What is Transfer Learning?

Transfer Learning is a machine learning technique where a model developed for one task is reused as the starting point for a model on a second, related task. In computer vision, this involves taking a model pre-trained on a massive dataset (like ImageNet) and adapting it to your specific task (e.g., classifying medical images).

The core benefit: It allows you to achieve high accuracy with a much smaller dataset and significantly less training time, because the model has already learned a rich set of general features (edges, textures, shapes).

12. Name a few transfer learning algorithms for image classification.

These are not algorithms but pre-trained model architectures famously used for transfer learning:

• VGG (VGG16, VGG19): Known for its simple, uniform architecture.
• ResNet (ResNet50, ResNet101): Introduced "residual connections" to enable much deeper networks.
• Inception (GoogLeNet): Uses "Inception modules" to capture features at multiple scales.
• MobileNet: Designed for high efficiency on mobile and embedded devices.
• EfficientNet: Achieves state-of-the-art accuracy by systematically scaling network dimensions.

13. Explain the VGG16 model in detail.

VGG16 is a deep CNN architecture known for its simplicity. The "16" refers to its 16 layers with learnable weights. Its defining feature is the exclusive use of small 3x3 convolutional filters stacked on top of each other. While powerful, it is a very large model (~138 million parameters) and computationally expensive compared to newer architectures.

14. How to finetune any pretrained model for different data?

Finetuning involves unfreezing the later layers of a pre-trained model and re-training them on your new dataset with a very low learning rate.

Steps:

1. Load a pre-trained model without its final classification layer (`include_top=False`).
2. Freeze all layers of the base model (`base_model.trainable = False`).
3. Add your own new classifier head on top.
4. Train the model so only the new layers are trained.
5. Unfreeze some of the top layers of the base model.
6. Re-compile and continue training with a very low learning rate (e.g., `1e-5`) to gently adapt the weights.

15. Why do we need to freeze layers in transfer learning and what will happen if we don't?

We freeze layers to preserve the valuable, generalized features that the model learned from the massive pre-training dataset. If we don't, the large, random gradients from the newly initialized classifier head will propagate backward and destroy the carefully learned weights, erasing the main benefit of transfer learning.

16. What is a residual network in the ResNet model?

A residual network is the core component of a ResNet model. It's a block of layers that implements a "skip connection" or "identity shortcut." Instead of learning a direct mapping, it learns a residual function and adds the original input back to the output. This combats the vanishing gradient problem and enables the training of extremely deep networks.

17. How to evaluate a model based on accuracy and loss graphs?

By plotting training and validation metrics over epochs, you can diagnose the model's behavior:

• Good Fit: Both training and validation loss decrease and converge.
• Overfitting: Training loss decreases, but validation loss flattens or increases.
• Underfitting: Both training and validation loss remain high and fail to improve.

Diagnosing model behavior by plotting training vs. validation metrics.

18. What is pickle and where it is used?

Pickle is a standard Python module for object serialization — converting a Python object into a byte stream. In machine learning, its primary use is to save and load trained models, label encoders, or any other Python object to disk, allowing you to reuse them without retraining.

19. What are callbacks and where are they used?

In Keras/TensorFlow, a callback is an object that can perform actions at various stages of the training process. They are used to automate and monitor training. Common callbacks include ModelCheckpoint, EarlyStopping, and ReduceLROnPlateau.

20. Which method of opencv is used to show an image?

The cv2.imshow() function is used to display an image in a window.

Remember that cv2.imshow() must be followed by cv2.waitKey() for the window to remain visible. A call to cv2.destroyAllWindows() is needed to close the windows cleanly.

21. What is model checkpointing?

Model checkpointing is the practice of saving the model's weights during training. It provides fault tolerance (to resume training if it crashes) and allows you to save only the best-performing model based on validation metrics, not just the last one.

22. How to save images in opencv?

The cv2.imwrite() function is used to save an image (NumPy array) to a file. The format is determined by the file extension.

23. How to understand the shape of the image in opencv?

Use the .shape attribute of the NumPy array. For a color image, it returns (height, width, channels). For a grayscale image, it returns (height, width).

24. How to process video in opencv?

Video processing involves a loop: create a cv2.VideoCapture object, read frames one by one using .read(), process each frame (which is just an image), display it, and break the loop on a key press. Finally, release the capture object.

25. What is image Normalisation in opencv and why it is needed?

Image Normalization is scaling pixel values to a standard range, typically [0, 1]. It's crucial for deep learning as it helps models converge faster and more stably. A common method is to divide the image array by 255.0.

26. Which algorithm is used to detect edges in video/images?

The most famous and effective is the Canny Edge Detection algorithm (cv2.Canny()). It is a multi-stage algorithm that excels at producing clean, thin edges while being less susceptible to noise.

27. What is OCR?

OCR (Optical Character Recognition) is the technology used to convert images of text into machine-encoded text, allowing a computer to "read" and process it.

28. How to extract text data from images?

The common method is to use an OCR engine like Tesseract. The workflow involves preprocessing the image with OpenCV (e.g., grayscaling, binarization) to improve accuracy, and then passing the clean image to the OCR engine.

29. How to blur the image in opencv and different types of blurring techniques?

Blurring (smoothing) reduces noise. Common techniques include Averaging (cv2.blur), Gaussian (cv2.GaussianBlur), Median (cv2.medianBlur), and Bilateral filtering (cv2.bilateralFilter), which preserves edges.

30. For what is the Sobel operation used in OpenCV?

The Sobel operation (cv2.Sobel()) is primarily used for calculating the image gradient, which is a measure of intensity change. This directly corresponds to edge detection.

31. What is a translation matrix and where it is used in opencv?

A translation matrix is a 2x3 matrix used to shift an image's position along the X and Y axes. It is used with the cv2.warpAffine() function to apply this geometric transformation.

32. How to rotate and translate the images in opencv?

Both are geometric transformations applied using cv2.warpAffine(). For rotation, you first generate the required 2x3 matrix using cv2.getRotationMatrix2D().

33. How to convert any rgb images to different types of image format like BGR, GRAY, HSV?

The cv2.cvtColor() function is the universal tool for all color space conversions, using flags like cv2.COLOR_BGR2GRAY and cv2.COLOR_BGR2HSV.

Remember that OpenCV's default color space is BGR. If you get an RGB image from another library (like Pillow), you must use cv2.COLOR_RGB2BGR to convert it.

34. How to convert image to gray scale?

Either load it directly as grayscale (cv2.imread('image.jpg', cv2.IMREAD_GRAYSCALE)) or convert a color image using cv2.cvtColor(image, cv2.COLOR_BGR2GRAY).

35. What's the difference between a 2D and 3D image?

This refers to the dimensionality of the NumPy array. A 2D image has a shape of (height, width) and represents a single-channel image (e.g., grayscale). A 3D image has a shape of (height, width, channels) and represents a multi-channel image (e.g., BGR color).

36. What is FPS in video?

FPS (Frames Per Second) is the frequency at which images (frames) are displayed. In computer vision, the processing FPS is a critical benchmark indicating how many frames your algorithm can analyze per second.

37. How to crop an Image?

Cropping in OpenCV is simply NumPy array slicing. The format is cropped = image[startY:endY, startX:endX].

38. How many algorithms are in OpenCV?

OpenCV contains over 2,500 optimized algorithms, and this number is constantly growing.

39. How many types of image filters are in OpenCV?

Filters can be categorized into: Smoothing/Blurring (Low-pass), Gradient/Edge Detection (High-pass), and Morphological Transformations.

40. What are Haar Cascade classifiers?

Haar Cascade classifiers are a classic machine learning-based object detection method, famous for being one of the first to achieve robust, real-time face detection. It uses Haar-like features and a cascade of classifiers to rapidly discard non-face regions.

41. How to detect faces in images and video?

The classic method is using a pre-trained Haar Cascade classifier. The modern, more accurate method is using a deep learning-based detector (e.g., from OpenCV's DNN module).

42. How to recognize unique faces in OpenCV?

Face recognition (identifying who) is more advanced than detection (finding a face). It involves detecting the face, then using a deep learning model to extract a unique feature vector called an "embedding", and finally comparing this embedding to a database of known faces.

43. What is a Histogram for images?

An image histogram is a graph representing the distribution of pixel intensities in an image. It provides a quick summary of the image's tonal range (dark, bright, high/low contrast).

44. How can you use a Histogram as an image feature?

A color histogram can serve as a simple but effective feature vector to describe the global color content of an image. You can compare images by calculating their histograms and measuring the similarity between them using cv2.compareHist().

45. What is an Image Search Engine and how to build it?

An Image Search Engine (Content-Based Image Retrieval) finds visually similar images from a database. To build one, you index the database by extracting a feature vector (e.g., a color histogram or a CNN embedding) for each image. To search, you extract the feature vector for a query image and find the closest vectors in your index.

46. Which method is used to draw a rectangle, line, circle, polygon on images?

OpenCV provides specific functions: cv2.line(), cv2.rectangle(), cv2.circle(), cv2.polylines(), and cv2.putText(). A `thickness=-1` fills the shape.

47. How can we Increase the Quality of an Image?

Techniques include: Noise Reduction (e.g., `cv2.bilateralFilter`), Sharpening (e.g., unsharp masking), Contrast Enhancement (e.g., `cv2.createCLAHE`), and Super-Resolution (using deep learning).

48. What is aspect ratio in images and why do we need to maintain it?

The aspect ratio is the ratio of an image's width to its height. It must be maintained during resizing to prevent the image from becoming distorted (stretched or squashed), which would harm the performance of most computer vision models.

49. What Is Image Transform?

An Image Transform is a geometric operation that modifies the spatial relationship of pixels. Common transforms include translation, rotation, scaling, affine, and perspective transforms.

50. What Do You Mean By Zooming Of Digital Images?

Zooming is enlarging an image. This requires creating new pixels to fill in the gaps through a process called interpolation (e.g., `cv2.INTER_CUBIC` or `cv2.INTER_LINEAR`).

51. Why do we need to resize and skip some images in video for Object Detection?

• Resizing: Object detection models require a fixed input size, and resizing reduces computation to enable real-time performance.
• Skipping Frames: If the detection algorithm is slower than the video's FPS, skipping frames is necessary to keep up with the live stream without causing significant lag.

52. What is Object Detection and which model is used to detect objects in real-time video?

Object Detection is the task of identifying and locating objects with a bounding box. For real-time video, models must be fast and accurate. The most popular family is YOLO (You Only Look Once), which is extremely fast because it processes the entire image in a single pass.

53. Define Resolution.

Resolution is the total number of pixels in an image, typically expressed as width × height (e.g., 1920 × 1080). Higher resolution means more detail.

54. What Is Image Translation And Scaling?

• Image Translation: Shifting the entire image's position without changing its size or orientation.
• Image Scaling: Changing the size of the image, either making it larger (zooming) or smaller (shrinking).

55. What Do You Mean By Shrinking Of Digital Images?

Shrinking is downscaling an image to reduce its size. This involves removing pixels. The recommended interpolation method in OpenCV for shrinking is cv2.INTER_AREA, as it properly handles combining pixel information to avoid artifacts.

---

Conclusion

Congratulations on making it through this comprehensive guide! By understanding these core concepts—from fundamental operations to advanced deep learning techniques—you are now well-equipped to tackle real-world computer vision challenges and excel in your interviews. The key to mastery is practice. Keep building, keep experimenting, and never stop learning. Good luck!