Python 3, 144 127 bytes

This solution uses cv2's awesome image processing power. Despite cv's less awesome, super long and readable method names, it beats both other Python answers!

Golfed:

import cv2,numpy as n
f=lambda b:n.amax(cv2.connectedComponents(b*255,0,4)[1])
def g(a):b=n.array(a,n.uint8);print(f(1-b),f(b))

Expanded:

import cv2
import numpy as np

# Finds the number of connected 1 regions 
def get_components(binary_map):
    _, labels = cv2.connectedComponents(binary_map*255, connectivity=4) # default connectivity is 8
    # labels is a 2d array of the binary map but with 0, 1, 2, etc. marking the connected regions
    components = np.amax(labels)
    return components

# Takes a 2d array of 0s and 1s and returns the number of connected regions
def solve(array): 
    binary_map = np.array(input_map, dtype=np.uint8)
    black_regions = get_components(1 - binary_map) # 0s
    white_regions = get_components(binary_map) # 1s
    return (black_regions, white_regions)