Accurate delineation of agricultural fields from remote sensing imagery is crucial for various precision agriculture and remote sensing applications. However, common learning-based delineation methods face problems related to limited sample availability and weak transferability. Conversely, nonlearning-based methods, while not reliant on training samples, face significant challenges with ineffective boundary detection and binarization, extensive boundary breaks, and inaccurate field object distinction. This study addresses these capability gaps with a novel integrated delineation method. The proposed method initially identifies preliminary field boundaries by leveraging texture variances and a modified local binarization method. Subsequently, a designed boundary optimization method is employed to repair broken boundaries and eliminate dangled boundaries. The final results are obtained after distinguishing field objects using our object-based cropland similarity indicator. The method was evaluated in six study areas in China and showed robust performance across varied agricultural landscape, with average boundary and area F1 scores of 0.849 and 0.826, respectively. Comparative experiments at step levels affirm that 1) our texture-based method provides clearer detection, and binarization method balances boundary preservation and noise suppression, yielding more well-defined boundaries, 2) the designed boundary optimization method effectively repairs long-distance boundary breaks and remove dangled boundaries, ensuring boundary-continuous and shape-regular results, and 3) the constructed object-level cropland similar index robustly and conveniently distinguishes field objects. Result-level comparisons highlight its superiority over existing nonlearning-based approaches, while comparisons with recent deep learning models affirm its value in certain scenarios. Overall, the proposed sample-free delineation method seamlessly integrates efficient and comprehensive methods, offering a high-quality field delineation solution without samples, and establishes a generalized framework of benchmark data production for extensive parcel-level agricultural applications.