Xiaobai Liu

81 papers A* 27B 4Misc 2Journal 42Unranked 5
YearRankTypeTitle / Venue / Authors
2024 J jnl
ACM Trans. Multim. Comput. Commun. Appl.
Huijie Zhang, Pu Li, Xiaobai Liu, Xianfeng Terry Yang, Li An
2024 conf
MMAsia
Huijie Zhang, Xiaobai Liu
2024 conf
MMAsia
Pu Li, Yibiao Zhao, Xiaobai Liu
2023 J jnl
CoRR
Pu Li, Xiaobai Liu
2023 J jnl
IEEE Trans. Multim.
Pu Li, Marie A. Roch, Holger Klinck, Erica Fleishman, Douglas Gillespie, Eva-Marie Nosal, Yu Shiu, Xiaobai Liu
2023 J jnl
CoRR
Pu Li, Marie A. Roch, Holger Klinck, Erica Fleishman, Douglas Gillespie, Eva-Marie Nosal, Yu Shiu, Xiaobai Liu
2023 J jnl
IEEE Trans. Circuits Syst. Video Technol.
Hua Bao, Ping Shu, Hongchao Zhang, Xiaobai Liu
2022 J jnl
IEEE Trans. Pattern Anal. Mach. Intell.
Hongjun Wang, Guanbin Li, Xiaobai Liu, Liang Lin
2022 J jnl
IEEE Trans. Pattern Anal. Mach. Intell.
Yuanlu Xu, Wenguan Wang, Tengyu Liu, Xiaobai Liu, Jianwen Xie, Song-Chun Zhu
2021 J jnl
Remote. Sens.
Huijie Zhang, Li An, Vena W. Chu, Douglas A. Stow, Xiaobai Liu, Qinghua Ding
2021 A* conf
ACM Multimedia
Pu Li, Xiaobai Liu, Xiaohui Xie
2020 J jnl
CoRR
Hongjun Wang, Guanbin Li, Xiaobai Liu, Liang Lin
2020 J jnl
IEEE Trans. Cybern.
Yudong Han, Lei Zhu, Zhiyong Cheng, Jingjing Li, Xiaobai Liu
2020 B conf
IJCNN
Pu Li, Xiaobai Liu, K. J. Palmer, Erica Fleishman, Douglas Gillespie, Eva-Marie Nosal, Yu Shiu, Holger Klinck, Danielle Cholewiak, Tyler A. Helble, Marie A. Roch
2020 J jnl
CoRR
Pu Li, Xiaobai Liu, K. J. Palmer, Erica Fleishman, Douglas Gillespie, Eva-Marie Nosal, Yu Shiu, Holger Klinck, Danielle Cholewiak, Tyler A. Helble, Marie A. Roch
2020 B conf
ICPR
Pu Li, Xiaobai Liu
2020 J jnl
IEEE Trans. Neural Networks Learn. Syst.
Xiaobai Liu, Qian Xu, Grayson Adkins, Eric Medwedeff, Liang Lin, Shuicheng Yan
2019 J jnl
CoRR
Yudong Han, Lei Zhu, Zhiyong Cheng, Jingjing Li, Xiaobai Liu
2019 J jnl
IEEE Trans. Multim.
Tianshui Chen, Riquan Chen, Lin Nie, Xiaonan Luo, Xiaobai Liu, Liang Lin
2019 J jnl
IEEE Trans. Circuits Syst. Video Technol.
Xiaobai Liu, Qian Xu, Thuan Chau, Yadong Mu, Lei Zhu, Shuicheng Yan
2018 A* conf
CVPR
Yuanlu Xu, Lei Qin, Xiaobai Liu, Jianwen Xie, Song-Chun Zhu
2018 J jnl
IEEE Trans. Circuits Syst. Video Technol.
Xiaobai Liu, Yuanlu Xu, Lei Zhu, Yadong Mu
2018 J jnl
Future Gener. Comput. Syst.
Wenbin Jiang, Min Long, Laurence T. Yang, Xiaobai Liu, Hai Jin, Alan L. Yuille, Ye Chi
2018 J jnl
ACM Trans. Intell. Syst. Technol.
Xiaobai Liu, Qian Xu, Yadong Mu, Jiadi Yang, Liang Lin, Shuicheng Yan
2018 J jnl
Pattern Recognit.
Kunqian Li, Wenbing Tao, Xiaobai Liu, Liman Liu
2018 J jnl
IEEE Trans. Knowl. Data Eng.
Xiaobai Liu, Qian Xu, Jingjie Yang, Jacob Thalman, Shuicheng Yan, Jiebo Luo
2018 A* conf
AAAI
Haoshu Fang, Yuanlu Xu, Wenguan Wang, Xiaobai Liu, Song-Chun Zhu
2018 J jnl
CoRR
Tianshui Chen, Riquan Chen, Lin Nie, Xiaonan Luo, Xiaobai Liu, Liang Lin
2018 J jnl
IEEE Trans. Pattern Anal. Mach. Intell.
Xiaobai Liu, Yibiao Zhao, Song-Chun Zhu
2018 A* conf
IJCAI
Xiaobai Liu, Donovan Lo, Chau Thuan
2017 J jnl
CoRR
Lei Qin, Yuanlu Xu, Xiaobai Liu, Song-Chun Zhu
2017 A* conf
AAAI
Yuanlu Xu, Xiaobai Liu, Lei Qin, Song-Chun Zhu
2017 J jnl
CoRR
Lei Zhu, Zi Huang, Xiaobai Liu, Xiangnan He, Jingkuan Song, Xiaofang Zhou
2017 J jnl
IEEE Trans. Multim.
Lei Zhu, Zi Huang, Xiaobai Liu, Xiangnan He, Jiande Sun, Xiaofang Zhou
2017 J jnl
CoRR
Haoshu Fang, Yuanlu Xu, Wenguan Wang, Xiaobai Liu, Song-Chun Zhu
2017 A* conf
ACM Multimedia
Xiaobai Liu, Qi Chen, Lei Zhu, Yuanlu Xu, Liang Lin
2017 ed.
VSCC@MM
Xiaobai Liu, Yadong Mu, Yu-Gang Jiang, Jiebo Luo
2017 A* conf
IJCAI
Chengcheng Yu, Xiaobai Liu, Song-Chun Zhu
2017 J jnl
IEEE Trans. Knowl. Data Eng.
Yadong Mu, Wei Liu, Xiaobai Liu, Wei Fan
2017 A* conf
ACM Multimedia
Xiaobai Liu, Yadong Mu, Yu-Gang Jiang, Jiebo Luo
2017 conf
VSCC@MM
Qian Xu, Xiaobai Liu
2016 A* conf
IJCAI
Xiaobai Liu, Yadong Mu, Liang Lin
2016 A* conf
IJCAI
Zhanglin Peng, Ruimao Zhang, Xiaodan Liang, Xiaobai Liu, Liang Lin
2016 J jnl
CoRR
Zhanglin Peng, Ruimao Zhang, Xiaodan Liang, Xiaobai Liu, Liang Lin
2016 J jnl
IEEE Trans. Image Process.
Chenglong Li, Hui Cheng, Shiyi Hu, Xiaobai Liu, Jin Tang, Liang Lin
2016 A* conf
IJCAI
Lei Zhu, Jialie Shen, Xiaobai Liu, Liang Xie, Liqiang Nie
2016 J jnl
IEEE Trans. Neural Networks Learn. Syst.
Ping Luo, Liang Lin, Xiaobai Liu
2016 A* conf
AAAI
Xiaobai Liu
2016 A* conf
CVPR
Yuanlu Xu, Xiaobai Liu, Yang Liu, Song-Chun Zhu
2016 A* conf
ACM Multimedia
Xiaobai Liu
2015 J jnl
CoRR
Yuanlu Xu, Liang Lin, Wei-Shi Zheng, Xiaobai Liu
2014 A* conf
CVPR
Xianjie Chen, Roozbeh Mottaghi, Xiaobai Liu, Sanja Fidler, Raquel Urtasun, Alan L. Yuille
2014 J jnl
CoRR
Xianjie Chen, Roozbeh Mottaghi, Xiaobai Liu, Sanja Fidler, Raquel Urtasun, Alan L. Yuille
2014 J jnl
IEEE Trans. Image Process.
Xiaobai Liu, Qian Xu, Jiayi Ma, Hai Jin, Yanduo Zhang
2014 J jnl
IEEE Trans. Image Process.
Xiaobai Liu, Qian Xu, Shuicheng Yan, Gang Wang, Hai Jin, Seong-Whan Lee
2014 A* conf
CVPR
Xiaobai Liu, Yibiao Zhao, Song-Chun Zhu
2014 A* conf
CVPR
Roozbeh Mottaghi, Xianjie Chen, Xiaobai Liu, Nam-Gyu Cho, Seong-Whan Lee, Sanja Fidler, Raquel Urtasun, Alan L. Yuille
2013 A* conf
ICCV
Yuanlu Xu, Liang Lin, Wei-Shi Zheng, Xiaobai Liu
2013 A* conf
CVPR
Xiaobai Liu, Liang Lin, Alan L. Yuille
2012 J jnl
ACM Trans. Multim. Comput. Commun. Appl.
Xiaobai Liu, Shuicheng Yan, Tat-Seng Chua, Hai Jin
2012 J jnl
ACM Trans. Multim. Comput. Commun. Appl.
Xiaobai Liu, Shuicheng Yan, Bin Cheng, Jinhui Tang, Tat-Seng Chua, Hai Jin
2012 J jnl
Pattern Recognit.
Liang Lin, Xiaobai Liu, Shaowu Peng, Hongyang Chao, Yongtian Wang, Bo Jiang
2012 J jnl
IEEE Trans. Image Process.
Xiao-Tong Yuan, Xiaobai Liu, Shuicheng Yan
2011 J jnl
IEEE Trans. Circuits Syst. Video Technol.
Xiaobai Liu, Liang Lin, Shuicheng Yan, Hai Jin, Wenbin Jiang
2011 J jnl
IEEE Trans. Circuits Syst. Video Technol.
Xiaobai Liu, Liang Lin, Shuicheng Yan, Hai Jin, Wenbing Tao
2011 A* conf
ICCV
Xiaobai Liu, Xiaotong Yuan, Shuicheng Yan, Hai Jin
2011 A* conf
CVPR
Xiaobai Liu, Jiashi Feng, Shuicheng Yan, Liang Lin, Hai Jin
2010 A* conf
ACM Multimedia
Xiaobai Liu, Jiashi Feng, Shuicheng Yan, Hai Jin
2010 J jnl
IEEE Trans. Pattern Anal. Mach. Intell.
Liang Lin, Xiaobai Liu, Song Chun Zhu
2010 Misc conf
ICASSP
Xiaobai Liu, Haifeng Gong, Shuicheng Yan, Hai Jin
2010 A* conf
CVPR
Xiaobai Liu, Shuicheng Yan, Jiebo Luo, Jinhui Tang, ZhongYang Huang, Hai Jin
2010 J jnl
IEEE Trans. Image Process.
Xiaobai Liu, Shuicheng Yan, Hai Jin
2009 conf
ROBIO
Jian'an Xu, Xiaobai Liu, Dinghui Chu, Lining Sun, Mingjun Zhang
2009 Misc conf
SAC
Xiaoying Sha, Xiaobai Liu, Jianting Wen
2009 A* conf
ACM Multimedia
Xiaobai Liu, Bin Cheng, Shuicheng Yan, Jinhui Tang, Tat-Seng Chua, Hai Jin
2009 A* conf
CVPR
Liang Lin, Kun Zeng, Xiaobai Liu, Song Chun Zhu
2009 A* conf
CVPR
Xiaobai Liu, Liang Lin, Song Chun Zhu, Hai Jin
2009 A* conf
ICDM
Xiaobai Liu, Shuicheng Yan, Jun Yan, Hai Jin
2008 B conf
ICPR
Xiaobai Liu, Liang Lin, Hongwei Li, Hai Jin, Wenbing Tao
2008 B conf
ICPR
Hongwei Li, Liang Lin, Tianfu Wu, Xiaobai Liu, Lanfang Dong
2008 conf
PACIIA (2)
Mingjun Zhang, Xiaobai Liu, Dinghui Chu, Shaobo Guo
redb/extractors/decompiler/bninja/decompiler.py
← Index redb/extractors/decompiler/bninja/decompiler.py python
import os
import time
import json

from .analysis.medium_level import MediumLevelAnalysis

# disable the plugins set by user for binary ninja
os.environ["BN_DISABLE_USER_PLUGINS"] = "True"
import traceback

# Binary Ninja imports (conditional)
try:
    import binaryninja
    from binaryninja import mainthread, Symbol
    from binaryninja.enums import SymbolType
    BINARYNINJA_AVAILABLE = True
except Exception:
    BINARYNINJA_AVAILABLE = False
    binaryninja = None
    mainthread = None
    Symbol = None
    SymbolType = None

# Support both package and standalone imports
try:
    # Package import (when imported from redb)
    from .utils.hashes import calculate_sha256, calculate_tlsh
    from .utils.license import set_license
    from .utils.logging import setup_default_logger
    from .analysis.strings import StringAnalysis

    # Only import modules that depend on Binary Ninja when available
    if BINARYNINJA_AVAILABLE:
        from .analysis.cfg import CFGAnalysis
        from .analysis.disassembly import DisassemblyAnalysis
        from .analysis.low_level import LowLevelAnalysis
        from .arch.creator import ArchitectureCreator
        from .custom_options import register_custom_analysis_options
        from .function_type import FunctionTypeAnalysis, FunctionType
        from .analysis.scores import ObfuscationScores

except ImportError:
    # Fallback to absolute imports (for multiprocessing spawned processes)
    from redb.extractors.decompiler.bninja.utils.hashes import calculate_sha256, calculate_tlsh
    from redb.extractors.decompiler.bninja.utils.license import set_license
    from redb.extractors.decompiler.bninja.utils.logging import setup_default_logger

    # Only import modules that depend on Binary Ninja when available
    if BINARYNINJA_AVAILABLE:
        from redb.extractors.decompiler.bninja.analysis.cfg import CFGAnalysis
        from redb.extractors.decompiler.bninja.analysis.disassembly import DisassemblyAnalysis
        from redb.extractors.decompiler.bninja.analysis.low_level import LowLevelAnalysis
        from redb.extractors.decompiler.bninja.arch.creator import ArchitectureCreator
        from redb.extractors.decompiler.bninja.custom_options import register_custom_analysis_options
        from redb.extractors.decompiler.bninja.function_type import FunctionTypeAnalysis


class BinaryNinjaDecompiler:
    """A Binary Ninja-based decompiler that replicates the functionality of GhidraDecompilerScript.
    This class extracts decompiled code, disassembly with multiple normalization levels,
    and control flow graph information from binary files.
    """

    MIN_FUNCTION_SIZE = 10  # instructions
    MIN_BLOCK_SIZE = 4  # instructions
    INVALID_STACK_SIZE = -1

    def __init__(
        self,
        filepath,
        timeout,
        log=None,
        exporters=None,
        index_prefix=None,
        filetype=None,
        goresym=None,
        decompile_modules=None,
    ):
        """Initialize the Binary Ninja decompiler.

        Args:
            filepath: Path to the binary file to analyze
            log: Logger object (optional)
            timeout: Maximum time in seconds for analysis (default: 1200)
            exporters: List of exporters for the results (optional)
            index_prefix: Prefix for elastic index (optional)
            filetype: Type of the file (optional)

        """
        self.filepath = filepath
        self.log = log if log else setup_default_logger("BninjaDecompiler")
        self.BNINJA_TIMEOUT = timeout
        self.bv = None
        self.analysis_results = None
        self.errors = []
        self.exporters = exporters
        self.index_prefix = index_prefix
        self.filetype = filetype
        self.goresym = goresym
        self.decompile_modules = decompile_modules or {"all"}

        set_license(binaryninja)

        # Map to track instruction categorization
        mainthread.set_worker_thread_count(3)
        register_custom_analysis_options(binaryninja)

    def log_error(
        self, message, function_name, address, exception=None, error_location="unknown"
    ):
        """Log an error during processing."""
        error_msg = f"Error in function {function_name} at {address}: {message}"
        if exception:
            error_msg += f" - {str(exception)}"

        self.log.error(error_msg)

        # Add to errors list
        error = {
            "function_name": function_name,
            "function_address": str(address),
            "error_location": error_location,
            "error_message": message,
            "error_details": str(exception) if exception else "",
            "error_type": type(exception).__name__ if exception else "Unknown",
            "timestamp": int(time.time() * 1000),
        }
        self.errors.append(error)

    def __enter__(self):
        """Context manager entry point."""
        self.log.info(f"Opening binary file: {self.filepath}")
        binaryninja.BinaryViewType.add_binaryview_initial_analysis_completion_event(
            self.on_analysis_complete
        )

        #self.bv = binaryninja.load(self.filepath, update_analysis=False)
        self.bv = binaryninja.load(self.filepath, update_analysis=True)
        if self.bv is None:
            raise ValueError(f"Failed to open file: {self.filepath}")

        self.log.info("Waiting for analysis to complete...")

        self.log.debug(f"Binja analysis complete: {len(list(self.bv.functions))} functions")

        # set the architecture
        # todo: personalize this for other architectures
        self.arch = ArchitectureCreator("x86").get()

        # apply goresym
        if self.goresym is not None:
            self.__apply_goresym()
        return self

    def on_analysis_complete(self, bv):
        # Request an additional update after analysis is complete to ensure IL generation
        self.bv = bv
        self.bv.update_analysis()

        return

    def __apply_goresym(self):
        file = self.goresym
        data = None
        try:
            data = json.loads(open(file, 'r').read())
        except Exception as e:
            self.log_error(
                "Failed to open file from goresym: ",
                file,
                e,
                "analyze_binary",
            )

        if data is None:
            return

        self.bv.begin_undo_actions()
        if data.get('UserFunctions') is not None:
            user_functions = data['UserFunctions']
            for func in user_functions:
                try:
                    start = int(func['Start'])
                    name = func['FullName']
                    if self.bv.get_function_at(start) is None:
                        self.bv.create_user_function(start)

                    sym = Symbol(SymbolType.FunctionSymbol, start, name, name, name)
                    self.bv.define_user_symbol(sym)
                except Exception as e:
                    self.log.warning(f"Failed to apply GoReSym symbol for UserFunction {func.get('FullName', 'unknown')} at {func.get('Start', 'unknown')}: {e}")

        if data.get('StdFunctions') is not None:
            standard_functions = data['StdFunctions']
            for func in standard_functions:
                try:
                    start = int(func['Start'])
                    name = func['FullName']
                    if self.bv.get_function_at(start) is None:
                        self.bv.create_user_function(start)

                    sym = Symbol(SymbolType.FunctionSymbol, start, name, name, name)
                    self.bv.define_user_symbol(sym)
                except Exception as e:
                    self.log.warning(f"Failed to apply GoReSym symbol for StdFunction {func.get('FullName', 'unknown')} at {func.get('Start', 'unknown')}: {e}")

        self.bv.commit_undo_actions()
        return

    def __exit__(self, exc_type, exc_val, exc_tb):
        """Context manager exit point - clean up resources."""
        if self.bv:
            # Make sure to cancel any pending analysis
            # (if we have no pending analysis, binary ninja will log an error)

            # todo(@nicolo): investigate
            # if hasattr(self.bv, "abort_analysis"):
            #    self.bv.abort_analysis()
            self.bv.file.close()

        self.log.info("Cleanup completed successfully")
        return

    def tag(self):
        """Return the tag for this extractor."""
        return "DECOMPILED"

    def _module_selected(self, module_name):
        """Check if a decompiler sub-module is selected."""
        return "all" in self.decompile_modules or module_name in self.decompile_modules

    def analyze_binary(self):
        """Run Binary Ninja analysis and return results.

        Respects self.decompile_modules to selectively run/skip sub-modules:
        - strings: independent, skipped if not selected
        - decompilation: leaf module, skipped if not selected
        - disassembly: always runs (backbone — provides hash linkage for all others)
        - llil: leaf module, skipped if not selected
        - cfg: leaf module, skipped if not selected

        Only selected modules' results are appended to the results dict for DB insertion.
        Disassembly is always computed for linkage but only inserted when selected.
        """
        try:
            results = {
                "decompiled": [],
                "disassembled": [],
                "cfg": [],
                "llil": [],
                "errors": [],
                "strings": [],
                "mlil": []
            }

            run_all = "all" in self.decompile_modules
            run_strings = run_all or "strings" in self.decompile_modules
            run_decompilation = run_all or "decompilation" in self.decompile_modules
            run_disassembly = run_all or "disassembly" in self.decompile_modules
            run_llil = run_all or "llil" in self.decompile_modules
            run_cfg = run_all or "cfg" in self.decompile_modules

            # Determine if we need the per-function loop at all
            need_per_function = run_decompilation or run_disassembly or run_llil or run_cfg

            #functions_list = list(filter(is_not_ext_lib_function, self.bv.functions))
            functions_list = list(filter(is_lib_or_thunk, self.bv.functions))
            functions_list = list(filter(self.is_too_few_blocks, functions_list))

            # Strings extraction — independent of per-function analysis
            if run_strings:
                results["strings"] = StringAnalysis(self.bv, functions_list).analyze()

            if not need_per_function:
                return results

            for function in functions_list:
                try:
                    # Decompilation (HLIL) — leaf module, skip if not selected
                    hlil_json = self.extract_hlil(function) if run_decompilation else None

                    # Disassembly — always compute (provides hash linkage for others)
                    disass_json = self.extract_disasm(function)

                    # CFG — leaf module, skip if not selected
                    cfg_json = self.extract_cfg(function) if run_cfg else None

                    # LLIL — leaf module, skip if not selected
                    lowlevel_json = self.extract_lowlevel(function) if run_llil else None

                    # we run mlil only if we have cfg
                    mlil_json = self.extract_mediumlevel(function) if run_llil else None

                    # Calculate fuzzy hashes for disassembly using utils.hashes
                    if disass_json:
                        disass_no_addr = disass_json.get("disassembled_function_no_addresses", "")
                        disass_json["tlsh_disassembly"] = calculate_tlsh(disass_no_addr)

                    if hlil_json and disass_json:
                        hlil_json["disassembled_function_hash"] = disass_json[
                            "disassembled_function_hash"
                        ]
                        disass_json["decompiled_function_hash"] = hlil_json[
                            "decompiled_function_hash"
                        ]

                        results["decompiled"].append(hlil_json)
                        if run_disassembly:
                            results["disassembled"].append(disass_json)

                    elif hlil_json:
                        hlil_json["disassembled_function_hash"] = None
                        results["decompiled"].append(hlil_json)
                    elif disass_json:
                        disass_json["decompiled_function_hash"] = None
                        if run_disassembly:
                            results["disassembled"].append(disass_json)

                    # Add bi-directional linkage between LLIL and disassembly with fuzzy hashes
                    # LLIL fuzzy hashes (tlsh_llil) are already calculated in lowlevel_json
                    if lowlevel_json and disass_json:
                        # Add disassembly info to LLIL
                        lowlevel_json["disassembled_function_hash"] = disass_json["disassembled_function_hash"]
                        lowlevel_json["tlsh_disassembly"] = disass_json.get("tlsh_disassembly")

                        # Add LLIL fuzzy hashes to disassembly for easy export access
                        disass_json["tlsh_llil"] = lowlevel_json.get("tlsh_llil")
                        disass_json["minhash"] = lowlevel_json.get("minhash")

                    elif lowlevel_json:
                        lowlevel_json["disassembled_function_hash"] = None
                        lowlevel_json["tlsh_disassembly"] = None
                    elif disass_json:
                        # No LLIL available for this disassembly
                        disass_json["tlsh_llil"] = None
                        disass_json["minhash"] = None

                    if lowlevel_json:
                        results["llil"].append(lowlevel_json)
                        results["mlil"].append(mlil_json)

                    # Add CFG linkage with disassembled_function_hash
                    # Also add cyclomatic_complexity to disass_json for similarity metrics export
                    if cfg_json and disass_json:
                        cfg_json["disassembled_function_hash"] = disass_json["disassembled_function_hash"]
                        disass_json["cyclomatic_complexity"] = cfg_json.get("cyclomatic_complexity")
                        results["cfg"].append(cfg_json)
                    elif cfg_json:
                        cfg_json["disassembled_function_hash"] = None
                        results["cfg"].append(cfg_json)

                    # Ensure cyclomatic_complexity is set even if no cfg_json
                    if disass_json and "cyclomatic_complexity" not in disass_json:
                        disass_json["cyclomatic_complexity"] = None

                except Exception as e:
                    self.log_error(
                        "Failed to process function: ",
                        function.name,
                        function.start,
                        e,
                        "analyze_binary",
                    )

            return results

        except Exception as e:
            self.log.error(f"Error in binary analysis: {str(e)}")
            return None

    def extract_mediumlevel(self, function):
        middle_level = MediumLevelAnalysis(function, self.bv, self.log)
        middle_level_result, errors = middle_level.analyze()

        for error in errors:
            self.errors.append(error)

        return middle_level_result


    def extract_lowlevel(self, function):
        low_level = LowLevelAnalysis(function, self.bv, self.log)
        disassembly_json, errors = low_level.analyze()

        for error in errors:
            self.errors.append(error)

        return disassembly_json

    def extract_cfg(self, function):
        try:
            llil = function.llil if hasattr(function, 'llil') else None
            cfg = CFGAnalysis(function, llil_function=llil).extract_function_cfg()
            return cfg
        except Exception as e:
            self.log_error(
                "Fatal error in CFG extraction",
                function.name,
                function.start,
                e,
                "extract_disassembly",
            )
            return None

    def extract_disasm(self, function):
        try:
            disass_analysis = DisassemblyAnalysis(
                self.arch, function, self.bv, self.log
            )

            # Create disassembly JSON
            disassembly_json, errors = disass_analysis.get_json()

            for error in errors:
                self.errors.append(error)

            return disassembly_json

        except Exception as e:
            self.log_error(
                "Fatal error in disassembly extraction",
                function.name,
                function.start,
                e,
                "extract_disassembly",
            )
            return None

    def extract_hlil(self, function):
        """Extract HLIL from a function."""
        try:
            # Access function.hlil directly - this will either return the HLIL or raise an exception
            # Removed hlil_if_available check as it was causing race conditions
            if function.hlil is None:
                return None

            if len(function.hlil.basic_blocks) == 0:
                return None

            # if function has one basic block, then compare the len of the instructions against minimum of our functions
            if len(function.hlil.basic_blocks) == 1:
                block = function.basic_blocks[0]
                if len(list(block.disassembly_text)) < self.MIN_FUNCTION_SIZE:
                    return None

            # Get decompiled code
            function_prototype = str(function)
            decompiled_code = str(function.hlil)
            if not decompiled_code or decompiled_code.strip() == "":
                self.log.warning(f"Empty decompilation result for {function.name}")
                return None

            callers = []
            for caller_site in function.caller_sites:
                if caller_site.hlil:
                    callers.append(str(caller_site.hlil))

            calls = []
            for call in function.call_sites:
                if call.hlil:
                    calls.append(str(call.hlil))

            analysis_score = ObfuscationScores(function.hlil)
            flattened_score = analysis_score.flattened_score()
            mba_score = analysis_score.MBA_score()

            if decompiled_code:
                # Create json object
                function_json = {
                    "decompiled_function_hash": calculate_sha256(decompiled_code),
                    "decompiled_function": decompiled_code,
                    "decompiled_function_name": function.name,
                    "decompiled_function_prototype": function_prototype,
                    "decompiled_function_address": function.start,
                    "function_type": FunctionTypeAnalysis(function)
                    .get_function_type()
                    .name,
                    "functions_caller": list(callers),
                    "functions_call": list(calls),
                    "flattened_score": flattened_score,
                    "mba_score": mba_score,
                }
                return function_json
            else:
                return None

        except Exception as e:
            self.log.warning(
                f"Failed to get HLIL for {function.name} at {function.start}: {str(e)}"
            )

        return None

    def extract(self) -> bool:
        """Extract and process all analysis results.

        Returns:
            bool: True if extraction was successful, False otherwise

        """
        self.log.info(f"Starting binary analysis on {self.filepath}")
        try:
            results = self.analyze_binary()
            if not results:
                self.log.error("Analysis failed to produce results")
                return False

            self.analysis_results = results
            self.log.info(
                f"Successfully analyzed binary: {len(results['decompiled'])} decompiled functions, "
                f"{len(results['disassembled'])} disassembled functions, "
                #f"{len(results['cfg'])} basic blocks, "
                f"{len(results['errors'])} errors"
            )
            return True

        except Exception as e:
            self.log.error(f"Error in extraction: {str(e)}")
            return False

    def is_too_few_blocks(self, function):
        if function is None:
            return False

        if function.basic_blocks is None:
            return False

        # when binary ninja does not wait for the analysis, it creates function stubs where basic blocks array
        # is not populated yet. Therefore, we disable this heuristic.
        #if len(function.basic_blocks) == 0:
        #    return False

        # if function has one basic block, then compare the len of the instructions against minimum of our functions
        if len(function.basic_blocks) == 1:
            block = function.basic_blocks[0]
            if len(list(block.disassembly_text)) < self.MIN_FUNCTION_SIZE:
                return False

        return True

def is_lib_or_thunk(function):
    function_type = FunctionTypeAnalysis(function).get_function_type()
    return not (function_type == FunctionType.THUNK or function_type == FunctionType.EXTERNAL or function_type == FunctionType.LIBRARY)