Image Border Padding with Zeros

Image Border Padding is a fundamental preprocessing operation in digital image processing and deep learning computer vision pipelines. This technique involves augmenting the boundaries of an image matrix by adding layers of specified values—typically zeros—around its perimeter.

In the context of grayscale images, each pixel is represented as an integer intensity value within a 2D matrix. When we apply border padding, we expand the spatial dimensions of this matrix by surrounding it with additional rows and columns filled with a constant value (zero in this case).

The Transformation: Given an image matrix of dimensions H × W (height × width) and a padding size P, the resulting padded image will have dimensions (H + 2P) × (W + 2P). The original pixel values remain intact in the center, while the newly added border regions are initialized to zero.

Mathematical Representation: For a padded output matrix O with original image I and padding size P:

$$O[i][j] = \begin{cases} I[i-P][j-P] & \text{if } P \leq i < H+P \text{ and } P \leq j < W+P \ 0 & \text{otherwise} \end{cases}$$

Why Zero Padding Matters: Zero padding is extensively utilized in convolutional neural networks (CNNs) to preserve spatial dimensions after convolution operations, prevent information loss at image boundaries, and maintain feature map sizes across network layers.

Your Task: Implement a function that adds zero-value border padding around a grayscale image matrix. Your implementation should:

• Add the specified number of rows/columns of zeros on all four sides (top, bottom, left, right)
• Return the padded image as a 2D list containing integer values
• Handle input validation and return -1 for any of the following error conditions:
  • The input is not a valid 2D array (non-rectangular or invalid structure)
  • The image has empty or zero dimensions
  • The padding size is not a non-negative integer