How does a CNN learn to recognize images?

• A. By manually coding image features
• B. By applying filters to identify patterns in images
• C. By analyzing pixel values directly
• D. By comparing images to pre-defined categories

Answer: (B)

A convolutional neural network (CNN) learns to recognize images primarily by applying filters to identify patterns within the images. This process involves layers of convolution where filters (or kernels) slide across the input image, detecting features such as edges, textures, shapes, and eventually more complex structures as the data moves through deeper layers of the network.

As the CNN processes the images, it captures spatial hierarchies in the data—starting from simple patterns in earlier layers to more complex patterns in later layers. The learning process is guided by backpropagation and optimization techniques, which adjust the weights of the filters based on the error in predictions compared to actual labels. This allows the CNN to become increasingly adept at distinguishing between different visual inputs.

In contrast to the other options, manually coding image features would be a labor-intensive process that lacks the adaptability and scalability of CNNs. Analyzing pixel values directly without processing my means of convolution leads to a loss of spatial relationships, reducing the effectiveness of pattern recognition. Lastly, while comparing images to pre-defined categories can be part of the classification stage, it does not capture the learning mechanism of a CNN, which primarily develops its understanding through feature extraction using filters.