Table 2.

Image segmentation Approaches with their advantages and drawbacks.

Methods	Strengths	Weakness
Thresholding	• Not require early information of image • Minimum computational difficulty • Beneficial to separate contextual and frontal properties	• Loss of intensity fluctuation • Difficult to arrange threshold • Dependent spatial features are removed • Ineffective for excessive edges
Edge Detection	• Human perception helps to create excellent contrast images. • Detect borders easily • Improves visual contrast on objects	• Sensitive to noise • Perform poorly with low-contrast • Not easy to create a closed curve • Low noise immunity compared to others
Region Based	• Improve noise immunity • Performs effectively with noise • Multiple criteria may be selected • Better complex shape handling	• Dual segmentation requires • Seed point selection is required • Region splitting results in segments • More computational cost
Clustering	• It shows data structures and patterns • Assist in choosing cluster-defining traits • Detects outliers and irregularities • Compresses picture data and saves storage	• Initial cluster centroids affect clustering • Outliers alter clustering • More computationally costly for large datasets • It requires database equality, which is not true for cluster densities
Semantic	• It is automated and saves time and effort • Versatile for not well-defined objects • Fine-grained characteristics and pixel-level object boundaries are discovered • Obtained higher accuracy than traditional methods	• Pixel-level categorization is computationally intensive and slows real-time systems • Labeling semantic segmentation datasets is expensive • Pixel-level labeling does not match real-world complexity
Instance	• Fine-grained localization uses pixel-level objects • Segment overlapping photos • Need extensive object data to detect size, shape, and location • Helps autonomous cars and robots recognize objects	• Computational intensity • Object and semantic segmentation are simpler than instance • Instance segmentation restricts applications • Complex instance segmentation models over segment related entities