# Introduction to Computer Vision (CV) #48 Apr 28, 2025 Multimodal https://www.aibutsimple.com/p/introduction-to-computer-vision-cv Computer vision (CV) is a highly active field of AI that trains computers to interpret and make use of visual information from data such as images and videos. It focuses on developing algorithms and models to process visual inputs and perform tasks that mimic human visual perception, like identifying objects, recognizing faces, or understanding an image’s context. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/be495173-409b-4d0e-b380-7a895bbbedd8/CV_illust__1_.png) Since 2010, computer vision has risen rapidly in popularity due to its ability to perform many tasks that computers were once thought incapable of. It is widely used across countless industries; here are some common applications: - In healthcare, computer vision helps diagnose diseases from medical imaging such as X-rays or CT scans. - Security systems rely on it for facial recognition, fraud detection, and ID verification. - In the automotive sector, it powers self-driving cars by detecting pedestrians and traffic signs. - The retail sector uses computer vision for inventory management and automated checkout. --- Big thanks to our partners for keeping this newsletter free. If you have a second, clicking the ad below helps us a ton—and who knows, you might find something you’ll enjoy. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/f2d5afd0-6dc9-4219-89a4-18f051fb8715/image.png) ### Learn AI in 5 minutes a day What’s the secret to staying ahead of the curve in the world of AI? Information. Luckily, you can join 1,000,000+ early adopters reading [The Rundown AI](https://subscribe.therundown.ai/?form=open) — the free newsletter that makes you smarter on AI with just a 5-minute read per day. [Sign up to start learning.](https://subscribe.therundown.ai/?form=open) --- ### Timeline Computer vision has evolved a lot during its research, spanning over six decades. Here’s a quick timeline: In 1959, researchers began exploring image processing by studying brain responses to visual stimuli, discovering that vision is hierarchical: it starts with simple shapes like edges and then progresses to create more detailed patterns. In the 1960s, AI emerges as a field, with early attempts to digitize and process images. Optical character recognition (OCR) was developed in the 1970s, allowing computers to read text and handwritten text using machine learning techniques (hard coding would be difficult and inconsistent). In the 1980s, there were many theoretical advancements, such as the vision hierarchy hypothesis (David Marr) and the Neocognitron introducing convolutional layers (Kunihiko Fukushima). ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/48abe720-785c-4506-a747-279403e3daf2/Copy_of_CV_illust.png) During the 1990s-2000s, standardized datasets improved convolutional neural network (CNN) training, leading to quick development, while real-time face recognition systems were introduced. In the 2010s, there was a deep learning revolution, fueled by large datasets (such as the ImageNet dataset) and CNN development (like the AlexNet in 2012), which greatly improved image model performance. ### Convolutional Neural Networks Computer vision primarily relies on one foundational model—the convolutional neural network (CNN). They are the backbone behind computer vision’s rise (and popularity) and its remarkable utility across many applications. CNNs learn hierarchical representations, starting with simple features (such as edges) in early layers and progressing to more complex ones (like eyes) in deeper layers. This ability to automatically learn features has made CNNs the gold standard for CV. Their architecture is intentional and intuitive for image tasks: - Convolutional layers process the image, detecting important patterns such as textures, edges, or shapes. They are hierarchical, identifying low-level features earlier and advancing to high-level features in deeper layers. - Pooling layers reduce the data’s spatial dimensions, reducing the number of parameters and compressing the data while maintaining as much information as possible. - Activation functions such as the relu are used to introduce nonlinearity, helping the model perform the target task accurately. - Finally, fully connected layers are used to perform the final target task, such as classification. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/6e45b7c6-f722-47e5-9897-f0fb093a6abd/image.png) As CNN research progressed, we developed standard architecture choices. The most common configuration uses a 3×3 filter size along with 2×2 pooling. Also, layers are stacked in “blocks,” using this specific order: convolutional layer, pooling layer, nonlinearity. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/f249cdfe-2154-47fe-8b66-ea057d25036e/image.png) Want to dive deeper into CNNs? Check out our past issues here: - [Introduction to CNNs, Simplified (Beginner)](https://www.aibutsimple.com/p/convolutional-neural-networks-simplified?utm_source=newsletter.aibutsimple.com&utm_medium=referral&utm_campaign=introduction-to-computer-vision-cv) - [CNNs, Mathematically Explained (Intermediate)](https://www.aibutsimple.com/p/convolutional-neural-networks-explained-mathematically?utm_source=newsletter.aibutsimple.com&utm_medium=referral&utm_campaign=introduction-to-computer-vision-cv) ### Computer Vision Tasks #### Image Processing In image processing, images are modified or information is extracted from them. The processing prepares images for use in models or for further analysis. Filtering (removing noise or blurring images to focus on key features), edge detection (identifying boundaries), and transformation (adjusting images by resizing, rotating, or warping) are common techniques. In the domain of transformation, a widely used image processing technique that not only increases the size of your dataset (large datasets are crucial for CV) but also increases performance is image augmentation. Image augmentation generates similar but distinct training examples after some random adjustments to the training images. These modifications allow models to rely less on certain features, improving generalization. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/414f1b6f-03a5-4bff-beeb-efbefea4dfc6/image.png) For instance, we may flip the image left and right or up and down, which usually does not change the category of the object. In addition, we can also randomly crop the image to make objects appear in different positions in the image at different scales, which can also reduce the sensitivity of a model to the target position. We can also change the image brightness, contrast, saturation, and hue. In practice, we will combine multiple image augmentation methods for the best generalizability: ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/ae2485e9-181c-45e8-8df0-1a17f2d084e2/image.png) #### Image Classification Image classification is an extremely popular CV task, and it involves classifying an image into a category based on its content. A classic example is the infamous MNIST handwritten digit classification problem, where a network attempts to predict a digit from 1 through 10 given a small 28×28 pixel handwritten digit image. - A more modern application is found in social media platforms, which use image classification to automatically detect and filter objectionable content uploaded by users. Typically, machine learning or deep learning models are trained to be classifiers. For simple image classification problems such as the MNIST, a CNN might not be needed, but for more complex problems, CNNs are absolutely needed to obtain a satisfactory image classification accuracy. Classification models use datasets of labeled images, such as the ImageNet dataset. Images are fed in, processed through the model’s layers, and classified based on the highest probability output. With CNNs having specialized layers for image processing (convolutional, pooling, etc.), the model gradually learns to recognize features like eyes or hands—this is how we make neural networks “see.” ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/f191f133-5697-4357-b255-1e5ccabacca4/image.png) How a CNN “sees” an image Additionally, by training a CNN for image classification, the model performs feature extraction automatically through the convolutional layers, removing the need to separately process features through a different method. --- ### Start learning AI in 2025 Keeping up with AI is hard – we get it! That’s why over 1M professionals read [Superhuman AI](https://www.superhuman.ai/c/confirmation?magiclink_subscription) to stay ahead. - Get daily AI news, tools, and tutorials - Learn new AI skills you can use at work in 3 mins a day - Become 10X more productive [Sign up and start mastering AI](https://www.superhuman.ai/c/confirmation?magiclink_subscription) --- #### Object Detection Another common task in CV used for many day-to-day applications is object detection. It powers self-driving cars by detecting cars and pedestrians, identifies defects on assembly lines, and supports a wide range of other systems. Object detection extends classification by not only identifying what objects are present but also locating them within the image—typically using bounding boxes. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/f0350bdf-880b-4a65-98ec-578f57d34c1b/Copy_of_Copy_of_CV_illust__600_x_600_px___2_.png) Models such as YOLO (You Only Look Once) or the R-CNN (Region-Based CNN) use CNNs with modifications to perform object detection with high accuracy. In object detection, when using a bounding box to describe the spatial location of an object, the bounding box is rectangular, determined by the x and y coordinates of the upper-left corner and the lower-right corner. In object detection, models are classified into one-stage (YOLO family) or two-stage (R-CNN family) detectors, with one-stage models detecting objects in one stage and two-stage models in two. Two-stage detectors first propose regions of interest (ROIs) and then classify them, while one-stage detectors such as YOLO perform detection in one go, making them faster but sometimes less accurate. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/f8378734-81c4-4ecb-a58e-0addd585aa8b/image.png) Sample YOLO model architecture Visualized above, these models build on CNN feature extraction (learning the edges, shapes, etc.), adding fully connected layers for localization to predict bounding boxes. #### Image Segmentation Image segmentation is a technique similar to object detection, but instead of using bounding boxes, we divide an image into segments or regions, where each segment is a specific object or area with similar qualities. This can be useful when we need exact pixel detections, for instance in medical imaging, where tumors and organs are segmented precisely in MRI scans. There are two main types of image segmentation—semantic and instance segmentation. Semantic segmentation assigns a class label to each pixel, grouping pixels into categories without distinguishing between instances of the same class. Instance segmentation builds on semantic segmentation by outlining the precise position of individual object instances, assigning unique labels to each (such as distinguishing between two dogs). ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/17f90d32-7515-4492-83bd-e24cfc631454/image.png) The U-Net, SegNet, and Mask R-CNN are popular segmentation models—they use special networks, which we call fully convolutional networks (FCNs). - In FCNs, the output is a segmentation map instead of a label. These models often use encoder-decoder architectures (like an [autoencoder](https://www.aibutsimple.com/p/deep-unsupervised-learning-autoencoders?utm_source=newsletter.aibutsimple.com&utm_medium=referral&utm_campaign=introduction-to-computer-vision-cv)), with the encoder extracting features (similar to CNNs) and the decoder upsampling the data to produce predictions. #### Facial Recognition Facial recognition is a specialized application of object detection and classification together, focusing on identifying individuals based on their facial features. In facial recognition, faces are detected first and then matched to a database. Nowadays, facial recognition is everywhere: we unlock phones with Face ID, security systems match faces in surveillance footage, and it is found in many authentication and security applications. ![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,width=1200,format=auto,onerror=redirect,quality=80/uploads/asset/file/3f1f17e3-0ce7-47ed-aa60-fc821c0a778c/image.png) To recognize a face, we start by detecting a face using object detection techniques, followed by feature extraction using CNNs to create a facial embedding, which is a vector representation of a face. These embeddings are then compared to a database using distance metrics (like cosine similarity) for identification. --- Here’s a special thanks to our [biggest supporters](https://www.aibutsimple.com/upgrade?utm_source=newsletter.aibutsimple.com&utm_medium=referral&utm_campaign=introduction-to-computer-vision-cv): Sushant Waidande --- If you enjoy our content, consider supporting us so we can keep doing what we do. Please share this with a friend! Want to reach 50000+ ML engineers? Let’s work together: [WORK WITH US 📈](https://www.passionfroot.me/aibutsimple?utm_source=newsletter.aibutsimple.com&utm_medium=referral&utm_campaign=introduction-to-computer-vision-cv) Feedback, inquiries? Send us an email at [edwin@aibutsimple.com](mailto:edwin@aibutsimple.com). If you like, you can also donate to our team to push out better newsletters every week! [Support us ☕](https://ko-fi.com/aibutsimple?utm_source=newsletter.aibutsimple.com&utm_medium=referral&utm_campaign=introduction-to-computer-vision-cv) --- That’s it for this week’s issue of AI, but simple. See you next week! —AI, but simple team