Edge detection is one of the most commonly used operations in the field of image processing. The traditional Canny edge detector is widely used in gray-scale image processing. However, this traditional edge detector is unable to deal with color images efficiently. The problem arises due to the double threshold technique: hysteresis threshold. In this paper, we have developed a new double threshold technique. Using this threshold, results of Canny edge detector in different color spaces like RGB, HSV and L*a*b* have been analyzed for accuracy and computational speed. Experiments show that the improved Canny edge detector gives better performance than the traditional Canny edge detector in all color spaces. Also the L* component of L*a*b* color space gives the best edge details with least computational cost in comparison to edge detection in all other color spaces.