This is the last and easiest step in the image recognition task. We provide the input data that contains both the training dataset and the test dataset. The training dataset has images tagged as cat and not cat. Once the model is trained, the test data, which contains images without tags, is processed and it identifies the closest resemblance to a cat from it.

There are a few major challenges in building an image recognition system and the most important one is the hardware issue. In the case of high definition images, the training set of few thousand images amounts to a few billion pixel values to be computed. The computation is not linear, but complex derivatives, which require a lot of computing power. To overcome these challenges, the following points should be taken care of:

• Use image compression tools that reduce the size of images without compromising on image quality
• Use grayscale and gradient versions of colored images
• Use a Graphical Processor Unit (GPU), which provides good computation power

Since we have understood the basics of how image recognition works, let's now perform the image recognition task. As mentioned in the introduction, we will use Amazon's S3, IAM, and Rekognition web services along with Raspberry Pi, its camera module, and an IR sensor.

Let's get into the details of each component of our project.