It is becoming increasingly essential to develop and implement K-12 artificial intelligence (AI) education initiatives as AI technologies permeate into and transform different aspects of society. Image recognition is a foundational AI concept due to its broad applications, making it important to explore effective ways to introduce it to children. In this work, we conducted a session on image recognition at an AI-focused summer camp for middle school students. We introduced concepts including computer representation of images, feature extraction, and model prediction. We evaluated AI-See, a novel web application we developed to teach kids about kernels and matrix multiplication. In this work, our research questions were: (RQ1) What evidence is there of student learning of aspects of image recognition?; (RQ2) How can software-based and unplugged activities complement each other?; and (RQ3) Can we design an engaging and fun session for teaching image recognition? Analysis of the collected quantitative and qualitative data showed that many students developed a strong understanding of different aspects of image recognition, including computer representation of images, feature maps, kernels, and matrix math. Two-proportion z-tests revealed statistically significant improvement in performance from pre- to post-survey on four out of five multiple choice items about feature maps and kernels, indicating effective learning on these concepts through our session.