Image recognition systems based on digital photos are widely used in many areas—from security and healthcare to tracking wildlife or identifying lost pets. This paper presents a practical approach that uses artificial intelligence to handle recognition tasks. We combine deep learning methods for feature extraction with a similarity-based classification technique. As a real-world example, we built a system to recognize lost or homeless dogs using a structured image database. We used a pre-trained deep learning model (VGG16) to extract visual features from dog images and then applied the K-Nearest Neighbors (KNN) algorithm to match these features with those in the database. The study discusses key aspects like how the image dataset was prepared, how images were preprocessed, and how different similarity measures (Euclidean, Manhattan, and Cosine) were used to compare image features. Our test results showed that combining deep learning for feature extraction with a simple classifier like KNN worked well, even when the image quality varied. This type of system can be used in many other areas where visual recognition is needed and can be improved further with real-time processing or custom-trained deep learning models.