Public concerns and official pressures around environmental protection as well as exhausting petroleum resources have brought about preferences in studying and applying environmental-friendly polymers instead of synthetic petroleum-based polymers. For this aim, biopolymers can have outstanding advantages in biodegradability by saving time, energy and effort spent on increasingly costs of polymeric wastes. Natural starch (being studied in this work), as one of the most abundant natural resources for polymer materials, is inexpensive and biodegradable. As starch necessarily needs for modifications and processing before being used as ideal green polymer material, it is vital to perform feature extraction and defect detection measures in structure by some method like image analysis. Fractal as a new geometry has circumstantially progressed recently in the fields of image processing, physical space–time, medical image analysis, electrochemical patterns, digital images, sounds etc. Box Count Fractal Dimension as a very important and popular part of fractal geometry can be a useful factor in feature extraction and pattern recognition. This paper presents a new method for defect detection in the structure of natural starch modification images using the fractal dimension (FD) along with mean and standard deviation of image color. This is performed via feature extraction based on artificial intelligence by MatLab R2013a for box counting algorithms. Otsu’s graythresh method by MatLab R2013a is applied to binarize the images. The results of proposed methodology are illustrated as log–log curves where the fractal dimensions are recognized by curve fitting (CF) tool with more than 95% accuracy. The outputs express that starch samples’ FD vary in the range from 1.636 to 1.926 among which the last is identified as non-defective polymer. Non-defective feature is of great importance for quality control measures and chemical reactivity being here highlighted as biodegradability