Object recognition represents currently one of the most developing and challenging areas of the Computer Vision. This work presents a systematic study of various relevant parameters and approaches allowing semi-automatic or automatic object detection, applied onto a study case of melons on the field to be counted. In addition it is of a cardinal interest to obtain the quantitative information about performance of the algorithm in terms of metrics the suitability whereof is determined by the final goal of the classification. Research will consist of texture analysis, color segmentation in the RGB and YCbCr color spectrums, and the combination of all extracted features. Classification methods such as manual threshold tuning and k-nearest neighbor will be used after extracting the necessary components to identify melons. Provided that the aforementioned approaches can be commonly described as feature-based, this work as aiming to cover solutions operating on both local and global scale subsequently continues by advanced techniques as for example the normalized spatial correlation based on a known sample of either texture of the whole object being sought.